1

THE UNION WAGE GAP AMONG VIETNAMESE SMES

By

Nina Torm*

Department of Economics

University of Copenhagen

MARCH 2011

ABSTRACT

Based on unique matched employer-employee panel data from 2007 and 2009, this paper examines

the union wage gap among small and medium enterprises in Vietnam, controlling for both firm and

worker characteristics. The results show that unionized firms have higher wages when non-wage

labour costs are included in the wage measure. Similarly, at the individual level, wages for union

members are higher than for non-members, yet only when comparing across both unionized and

non-unionized firms. However, within unionized firms union membership is strongly associated with

receiving social insurance, in accordance with the firm level results. Thus, Vietnamese trade unions

appear to play a particularly important role in terms of ensuring the provision of social benefits.

Moreover, the union wage differential seems to be mostly a feature of Southern provinces, where

firms are also more likely to comply with social insurance regulations.

JEL Classification: J51 , O17, O53

Keywords: Trade Unions, Wages, Vietnam

* Nina.Torm@econ.ku.dk is associated with the Development Economics Research Group (DERG) at the Department of Economics, University of Copenhagen. Address for correspondence: Øster Farimagsgade 5, building 26, DK-1353 Copenhagen K., Denmark. I am grateful for productive and stimulating collaboration with the survey teams from the Vietnamese Institute of Labor Science and Social Affairs (ILSSA) and staff at Central Institute for Economic Management (CIEM). Valuable comments and suggestions from Simon McCoy, John Rand, Christoffer Sonne-Schmidt and Finn Tarp are gratefully acknowledged. The usual caveats apply.

2

1. Introduction It is widely documented that unionized establishments pay higher wages than otherwise comparable

non-union firms, yet with some variation depending among other factors on the competitiveness of

the labour market, and the degree of centralization and coordination. The vast majority of the

enormous trade union literature focuses on OECD countries, whereas evidence from developing and

transition countries remains scarcer, due in great part to limited data availability. However, in many

of these countries the paucity of institutional support for workers means that trade unions represent

one of the few institutional mechanisms potentially capable of promoting some measure of equity

and social justice. For this reason, examining their role and impact in a development context is

interesting as well as worthwhile from both a research and a policy perspective.

In the case of Vietnam, the transition to a market economy has meant a significant change in the

institutional setting in which trade unions operate. State-owned enterprises (SOEs), traditionally the

stronghold of trade unions, have gradually been equitized resulting in a reduction in union members,

whereas domestic private companies including small businesses, and foreign owned firms have

become increasingly important employers, creating new bases for the establishment of local trade

unions (Edwards and Phan, 2008). National figures indicate that the share of unionized firms has

increased in recent years, and so has union membership (VGCL, 2010) suggesting that unionization

must be associated with certain benefits. Moreover, with the recent positive changes in

implementing Corporate Social Responsibility (CSR), unions have become even more prominent, and

the pressure on them to live up to their role in terms of monitoring the observance of labour

legislation and act in the interests of their members has increased.

It is within this context that this paper analyzes whether trade unions are associated with higher

wage outcomes, using firm level survey data of small and medium (non-state) manufacturing firms in

Vietnam from 2007 and 2009. The firm panel has been matched with employee data from a

subsample of workers in each year, resulting in a matched panel. The survey covers micro, small and

medium firms, yet the focus here is on small and medium firms, since the establishment of trade

3

unions is mandatory only for enterprises with more than 10 employees (Nguyen et al. 2006: 603). 1

Moreover, firms with more than 10 employees operate under the Enterprise Law and the Labour

Code and thus are mandated to provide social benefits for workers with contracts of more than 3

months (Labour Code, Article 141). The Labour Code provides the impetus for formalizing and

strengthening the role of unions, yet the “effectiveness” of workplace trade unions remains

questionable due among other factors to their marginal independence, employer dominance and

the generally low incidence of collective agreements. Thus, it is not it is not a given that unionized

firms should ceteris paribus pay higher wages compared with similar firms that are not unionized.

The advantage of using matched employer-employee data is that it allows for disentangling worker

from firm heterogeneity by controlling for firm and worker characteristics which could affect both

union status and wage outcomes. Union membership information allows for separately estimating

the individual wage gap, which may differ from the firm wage differential since not all workers

(although the vast majority) in unionized firms are union members. The results show that unionized

firms pay higher real wages, a finding that is especially robust when non-wage labour costs are

included in the wage measure, and is further supported by evidence of a positive association

between unionization and other labour costs. At the individual level basic wages for union members

are significantly higher than for non-union members, but not when restricting the sample to

unionized firms only. However, within unionized firms union membership appears to be associated

with the provision of social benefits, in accordance with the firm level results. Moreover, the union

wage gap is particularly strong in Southern provinces, which is supported by evidence that firms in

these provinces are more likely to comply with social insurance regulations.

In the following section I provide a selective overview of the existing literature and theoretical

background. Section three discusses the specific Vietnamese context followed by section four which

describes the data. Section five outlines the methodology and the variables included in the empirical

analysis and section six presents the results of the analysis. The main conclusion, as outlined in

section seven, is that Vietnamese unions seem to be particularly accountable towards their

members when it comes to ensuring the provision of non-wage benefits.

2. Literature and theory

1 The World Bank SME Department currently operates with three groups of SMEs: micro, small and medium scale firms. Micro-enterprises have between 1 and 10 employees, small-scale enterprises between 11 and 50 employees, and medium-size enterprises between 51 and 300 employees. These definitions are broadly accepted by the Vietnamese Government (see Government decree no. 90/2001/CP-ND on “Supporting for Development of Small and Medium Enterprises”). In what follows, I apply these definitions.

4

Following the seminal work by Freeman and Medoff (1984) “What do Unions do?”, numerous trade

union studies, especially on the US, the UK and other OECD countries have emerged (for an early

survey see Lewis, 1986). More recent work includes Booth (1995) and Boeri and van Ours (2008)

assessing the role of unions and analyzing the decline in union density observed in the US and

Western Europe around the 1980s. In general, the literature portrays unions as operating along at

least three dimensions: their wage-making practices, their participatory role in terms of negotiating

and administering the rules for hiring, training, promoting, disciplining, firing and laying off workers,

and their activities as a pressure group on government (Pencavel, 2005). In this paper, I focus on the

first of these.

From a theoretical perspective, the trade union-wage gap comes from the fact that trade unions

introduce a wedge between the reservation wage of the worker and the value of a job, or in other

words between the labour supply and labour demand schedules. 2

The ability of the union to achieve

a wage rate that is higher than the non-union level depends on various factors including the

bargaining power of the union (the power of the union to act as a monopolist in the supply of

labour) and the existence of economic rent or surplus in the product market. In a bargaining model

where all union members have the same preferences, wages will be higher the stronger the

bargaining power of unions, and the lower the responsiveness of labour demand and of profits to

wages. These relations in turn depend on the competitiveness of the product market. In a perfectly

competitive market the union imposed wedge is reduced since a firm facing an infinite elasticity of

demand is unable to pass any union wage differential on to prices. Thus, unions might be able to

capture quasi-rents from capital in the short-run, but in the long-run the firm would leave the

market in search of a higher return on capital or be forced out by non-union firms with lower costs.

The extent of the union-wage association also depends on the degree of coordination and

centralization of wage determination activities (Boeri and van Ours, 2008). As highlighted by

Freeman and Medoff (1984) unions provide employees with a mechanism to shape the employment

relationship and their working environment, serving in a sense as an agent for a firm’s employees. In

most OECD countries union coverage (the percentage of the eligible workforce whose contract is

regulated by a collective agreement) exceeds union density (Nickell and Layard, 1999), and collective

agreements ensure that non-members (whether firms or workers) are covered by the union

2 Booth (1995) and Boeri and van Ours (2008) provide comprehensive reviews of trade union theory.

5

negotiated wage rate. This possibility to free-ride on the services and benefits provided by the union

is one of the reasons behind the declining union density (membership) seen in many OECD countries

since the 1980s (Boeri and van Ours, 2008). And coverage is therefore often seen as a better

measure of the bargaining power of unions. Even in the absence of extension rules, the wages of

non-union workers are, however, likely to be affected by the presence of unions through the “threat

effect” of unionism whereby non-union employers, will pay their workers wages that are

comparable to unionized workers (Pencavel, 2005). In the absence of such a threat effect (but in the

presence of unions), workers disemployed from union firms due to a wage-push, will seek

employment in non-union firms, shifting the labour supply right and reducing wages in this sector

below what they would be in the absence of unionism. Therefore, whether a threat effect exists or

not, wage levels appear to be affected by the presence of unions.

Empirical work on unions has been mostly based on either firm level or individual data, yet since the

mid-1990’s a growing number of studies using matched employer-employee data have emerged3

. In

terms of wage outcomes, for instance, Lalonde et al. (1996) use American data to examine the

effects of newly formed unions on total output, employment, material purchases, wage rates, and

productivity and find that firms reduce the first three but do not experience higher wage rates. By

contrast, in the UK, Hildreth and Pudney (1997) find higher wages among union firms and

interestingly the union wage premium is higher for individuals covered by the collective agreement

but who are not union members (in the UK collective agreements apply to all workers in the covered

jobs regardless of the employees’ union status). This could perhaps be partly due to non-union

members being exempt from the union membership fee. Using matched employer-employee data

from Spain, Card and De La Rica (2006) find that firm-level contracting is associated with a 5-10 pct.

individual wage premium. Moreover, the wage gain is larger for more highly paid workers, which is

in contrast to UK and US studies where unions have tended to “flatten” wages across skill groups

(Lewis, 1986). This difference could, however, be related to the fact that in the UK and US the

reference group is the non-union sector, whereas in Spain it is the prevailing sectoral agreement.

As mentioned earlier, work on developing countries is more limited, and perhaps more varied. Rama

(2000) states that in developing countries union members usually make 5-30 pct. more than non-

unionized workers, yet finds (atypically) that in CFA countries union membership is associated with

wages that are between 8-12 pct. lower (this could be related to the weak nature of the labour

movement in these countries). In the case of South Africa, Schultz and Mwabu (1998) find evidence 3 See Abowd and Kramarz (1999) for a review of methods and results.

6

of wage flattening at the high end of the wage distribution, whereas at the tenth percentile there is

a 41 pct. union wage gap. In a case study of 10 Vietnamese firms, Clarke et al. (2007) document that

trade unions are able to negotiate wages that are 5 pct. higher than wages in non-union firms.

Moreover, Edwards and Phan (2008) argue that since trade unions in Vietnam are involved in the

central wage decision-making processes, wages would be lower were it not for the presence and

influence of trade unions, yet provide no evidence of this.

The main contribution of this paper is to empirically test whether trade unions are associated with

higher wages among Vietnamese SMEs. To set the scene, the next section provides an overview of

the nature of Vietnamese trade unions and the context in which they operate.

3. Trade unions in the Vietnamese context

Under central planning trade unions functioned mostly as a “transmission belt” of the ruling

communist party, as described in Zhu and Fahey (2000). Yet, with Vietnam’s transition to a market

economy a rights-based system for the regulation of industrial relations has been established

allocating a more prominent role to trade unions. As in other Asian countries Vietnamese unions

operate primarily at the enterprise level. According to the Labour Code (promulgated in 1994 and

subsequently amended in 2002, 2006 and 2007

Based on 2009 enterprise (register) data from GSO, figures from the Vietnam General Confederation

of Labour (VGCL) show a general increase in unionization from around 45 pct. of firms at the end of

2007 to around 50 pct. in mid-2010 (VGCL, 2010).

), all enterprises with 10 employees or more are

required to form unions within six months of establishment (Article 153). Moreover, employers are

responsible for recognizing and facilitating the establishment of trade unions, and creating a

conducive environment for their members to carry out activities (Article 154). Employers shall not

discriminate against workers who form trade unions or participate in trade union activities, and are

required to pay the full wage and to provide the same rights, benefits and privileges enjoyed by

other workers in the firm to workers who carry out union activities full time. Moreover, an employer

cannot unilaterally terminate an employee who is a member of the executive committee of the

trade union (Article 155).

4

4 Each firm (primary) trade union is subordinate to one of 623 district level organizations, which in turn are subordinate to the 64 provincial federations and 20 central sectoral unions that comprise the Vietnam General Confederation of Labour (VGCL) (Clarke et al., 2007).

This trend is largely attributed to the rise in

unionization among private firms, limited liability companies, joint stock firms and cooperatives,

whereas among SOEs there has been a decline in union density (as these firms are being equitized).

7

Despite this positive trend a large number of firms remain non-unionized, and there appears to be

very little pressure on employers to set up unions, nor any sanction prescribed by the law for failure

to comply (Clarke et al., 2007). The absence of unions has been described as a leading cause of the

surge in wildcat strikes during the mid-2000s providing the only option (apart from exit) for workers

to “voice” their demands for higher wages and other concerns. In order to address this, the Labour

Code chapter (IVX) on disputes and settlement of disputes and strikes was amended in 2006.

The Trade Union Law (1994, and not amended since) states that it is only necessary for a trade union

to “inform the government body or organization concerned in order that official relations may be

established” (Article 2.1), indicating that the enterprise trade union is free to independently

represent workers’ rights and interests, yet in practice a trade union can only be established legally

under the umbrella of the VGCL – the higher level trade union – which remains closely integrated

into the Party-state structure and collaborates closely with the local labour department under the

supervision of the relevant Party body. This subordination of the trade union to the Party constitutes

a serious structural flaw of Vietnamese trade unions, and means that workplace trade unions have

limited independence and ability to act as a pressure group on government. The Trade Union Law

does not set out specific requirements for wages to be above the minimum or average wage in a

particular industry, thus the wage level depends on the bargaining power of the different partners.

However, upon the formulation of a wage scales and labour rates, the employer must consult the

executive committee of the trade union of the enterprise (Article 57). And local trade unions are

expected to supervise firms’ implementation of minimum wage legislation.5

The negotiation and monitoring of collective agreements provides an important “test” of the

effectiveness of firm trade unions in representing the interests of their members, and collective

agreements should in principle include agreed wage and bonus scales.6 The Labour Code mandates

that enterprises with more than 10 employees and all foreign-invested enterprises should have

legally binding collective agreements (Article 44), yet it is far from all unionized firms that have such

agreements. For instance, Clarke et al., (2007) note that only 20 pct. of unionized private sector

firms have collective agreements, and that the figure is much higher in Ho Chi Minh City (around 65

pct., compared with only around 6 pct. in Hanoi). 7

5 For more see

According to ILSSA (2010) only the textile industry

http://vietnamnews.vnagency.com.vn/Social-Isssues/206630/Ministry-enforces-minimum-wage-law.html 6 Note that Vietnam has yet to ratify the international conventions related to freedom of association and collective bargaining set out by the International Labour Organization of the United Nations. 7 Compared with China, with which Vietnam shares many industrial relations features, the pace of collective bargaining developments in Viet Nam has been much slower. Reasons for this include the reluctance of the

8

has a collective bargaining system which ensures that the wages of those working in a hazardous

environment are higher than under normal working conditions.

The higher level unions provide workplace unions with a collective agreement model, yet there is no

specialized material for training in wage negotiations, and only 10-15 pct. of union officers receive

any training with priority given to higher level officials on the grounds that their tenure is more

stable (Clarke et al. 2007). Thus, local trade unions are very much left to their own devices, and their

performance depends to a large extent on the personality of the union leader (chair) who is chosen

by management and may in some cases be the relatives or friends of owners or managers. This

personal kinship, it is argued, prevents the union from representing workers and protecting workers’

interests effectively (Zhu and Fahey, 2000). Moreover, the control and dominance of employers and

lack of true freedom of association has been cited as one of the major reasons that workers do not

see the advantage of unions (Nguyen et al. 2006

).

There seems, however, to be an increasing awareness of the weaknesses of having exclusively

enterprise-based industrial relations institutions. For instance, Yoon (2009) provides evidence of

efforts to strengthen the role of trade unions at the upper levels to offer better support to workers

and unions at the enterprise level, and to develop bargaining arrangements at the supra-enterprise

level. For instance, a newly formed garment industry union is developing a pilot sectoral bargaining

scheme in consultation with the government and the Vietnam Garment Manufacturers’ Association.

In terms of union membership in the non-state sector this has increased from 26 pct. to 31 pct. of

the total unionized labour force between the end of 2007 and mid-2010 (VGCL, 2010), whereas in

the state sector there has been a decline.

Thus, although union membership is not compulsory it

seems that workers perceive union membership to be beneficial. This is also reflected in the

employee component of the survey on which this paper is based. Here respondents were asked

what the benefits of trade union membership were, and around 50 pct. replied that “it secures that

the firm pays social benefits”, followed by 18 pct. indicating “more job security”, 13 pct. saying

“more safety at the workplace and 9 pct. replying “better and more stable wages”.

employer to undertake obligations that he/she may not be able to fulfill and the perceived pointlessness of signing a collective agreement that contains no provisions other than the minimum legal standards. Moreover, by contrast to in China the collective bargaining system in Vietnam has been designed for enterprise-level bargaining with no legislative provisions to enable bargaining at levels other than the enterprise.

9

Thus, trade unions seem to be living up to their mandate of ensuring the protection of workers as

specified in the Labour Code8. Further, with the recent focus on CSR urging the creation of internal

rules to clarify the responsibilities and obligations of each party in the employment relationship

workplace trade unions have indeed become more prominent, and under increasing pressure to

fulfill their role in terms of monitoring the observance of labour legislation and act in the interests of

their members (Quang, 2008). Moreover, in recent years trade unions have been highly active in

organizing programmes to educate workers about labour and other Laws.9

However, given the low

incidence of collective bargaining, and the absence of specific requirements on wage levels, the

extent to which unions are associated with higher wages becomes an empirical question.

4. Data

Two quantitative SME surveys, conducted in 2007 and 2009, are used in this paper (see CIEM, DOE,

ILSSA, 2010 for details on the surveys). The surveys trace firms over time covering around 2,600

enterprises in ten provinces (Ho Chi Minh City (HCMC), Ha Noi, Hai Phong, Long An, Ha Tay, Quang

Nam, Phu Tho, Nghe An, Khanh Hoa and Lam Dong). In both years and in all the areas covered by the

surveys, samples were stratified by ownership form to ensure that all types of non-state enterprises,

including officially registered households, private firms, cooperatives, limited liability companies and

joint stock enterprises were represented. For reasons of implementation, the surveys were confined

to specific areas (districts) in each province/city. The sampling scheme of the SME survey is based on

a representative sample of registered household and non-household firms drawn from GSO

enterprise census information (GSO, 2004 and 2008), and a random selection of informal firms

within the selected survey districts observed by the enumerator.

In both 2007 and 2009 the surveys included a separate employee module, consisting of randomly

sampled employees from a random sub-sample of firms representative of the full firm sample along

the size, legal status, location and sector dimensions. In 2007, the employee module was carried out

in a sub-sample of 582 firms covering 1,043 employees and in 2009, the corresponding figures were

577 firms and 1,444 employees. Within each firm, at least one manager and two production workers

were interviewed. Thus, this analysis is based on a unique matched employer-employee panel

dataset, allowing for the inclusion of both firm and employee characteristics when examining the

8 According to Article 12, Unions shall look after and protect the rights of employees; and inspect and supervise the implementation of the provisions of the labour laws including in terms of social benefit provision. 9 See http://vietnamnews.vnagency.com.vn/Domestic-Press-Highlights/205347/Trade-unions-educate-workers-on-labour-law.html

10

union wage association. Moreover, having a panel of workers allows for taking account of any

worker fixed effect (as well as firm fixed effect, given that the worker is observed in the same firm

over two years), which if unaccounted for could lead to a biased estimate on the union wage

differential.

As mentioned earlier, the establishment of trade unions is mandatory only for enterprises with more

than 10 employees, and therefore this analysis focuses on firms that were classified as small or

medium sized (with more than 10 employees) in both 2007 and 2009. For enterprises with fewer

than 10 workers (micro firms), establishing a trade union is not economically viable, and since a

substantial share of these firms operate as informal household firms they are not subject to the

Labour Code, and thus not mandated to contribute to social insurance – an important component of

the total wage package. Since all informal firms fall under the household (micro) category, this

analysis considers only formally registered firms.10 After applying this selection criterion, and

undertaking a thorough data cleaning (see appendix for more detail) including checking consistency

of time-invariant variables between the two survey rounds, I was left with an unbalanced panel of

383 firms (from where employees had been sampled) and 188 in the balanced panel. As for the

employee sample this consists of 1153 workers (unbalanced panel) and 226 (113 unique

observations) in the balanced panel. 11

5. Empirical strategy and summary statistics

The relatively higher ratio of balanced to unbalanced firms

compared with workers indicates a higher turnover (attrition) rate among workers than firms.

The analysis is divided into two parts. First, I estimate a firm wage equation (i) using firm

characteristics only and second, I estimate an individual wage equation (ii) where wages depend on

both worker attributes and the firm characteristics where the worker is employed

More specifically, the first specification is as follows:

(i) ln j j j j jY Z X Uα γ η δ ε= + + + +

10 According to the Decree No. 88/2006/ND-CP of August 29th 2006 on business registration, when a firm has more than 10 employees and/or uses more than one business premise, it may no longer operate as a household firm (formal or informal), and should register as a company under the Enterprise Law (2000). 11 For the firm sample only 5 observations were dropped due to missing wage information, thus selection bias along this dimension does not seem to be a cause for concern. For the employee sample 22 observations were omitted following an outlier test and 33 were dropped due to missing wage information (see appendix for more detail). In order to rule out selection bias a Heckman two-step procedure was carried out using the share of managers and whether the worker has a permanent contract as exclusion restrictions. The results (available upon request) reveal an insignificant Mills ratio and no significant difference between the OLS and Heckit estimates, thus there seems to be no evidence of selection bias along the wage dimension.

11

where the log of (real) firm wages at firm j ( ln jY ) depend on: a vector of firm-level covariates ( jZ ),

the characteristics of all workers at the firm j ( jX ), an indicator for the presence of a firm union

( jU ), and a firm level error component jε .

The second specification takes the form:

(ii) ln ij ij j ij ijY X Z Uα β γ δ ε= + + + +

where the log of (real) individual wages for worker i ( ln ijY ) depend on: a set of individual

characteristics ( ijX ), a vector of firm-level covariates for the firm where worker i is employed ( jZ ),

and an indicator for whether the individual is member of a firm level trade union ( ijU ), and a worker

specific error term ( ijε ).

From a methodological point of view there are at least two major challenges with this kind of

analysis. First, omitted variable bias arising from union status (whether at the firm or individual

level) being systematically correlated with either unobserved firm specific wage factors or

unobserved worker ability (in the case of second specification). In order to deal with this, I control

for a number of covariates (as described in the summary statistics below). Further, in the second

specification, I make use of the variation in individual union status in order to estimate a fixed

effects model. A benefit of observing individuals changing union status whilst staying within the

same firm (aside from the fact that the firm-specific effect remains constant) is that union and non-

union members are more comparable than sets of workers where the difference in union status

arises from the fact that workers move to a different firm. In the latter case, since “movers” may be

very different to “stayers”, for instance if movers are motivated by higher wages, this may introduce

selectivity bias (see Booth, 1995: 176-177).

Second, selection bias arising from the fact that (i) firms choose to become unionized and (ii) given

the existence of a firm union, individuals select into union membership if they believe this to be

beneficial. This self-selection of workers may in turn generate a positive correlation between wages

and union membership that is due to reverse causality: it is because wages are higher in these jobs

or occupations that workers are unionized rather than the other way around.12

12 To address this endogeneity problem and rule out selection bias arising from it, I use a Heckman two-step procedure. Further detail is provided in section 6.

Moreover, this “two-

step selection” has to be kept in mind when interpreting the results, since the fact that non-union

members who work in non-union firms do not have the option to be union members means that in

12

specification (ii) any observed difference in wages may not reflect the individual wage gain

associated with union membership, but rather the wage differential between being employed in a

union versus a non-union firm. However, the fact that not all firms have unions combined with

there being non-members in unionized firms, allows for estimating a “pure” individual wage

premium by restricting the sample to unionized firms. This is what distinguishes the current analysis

from trade union studies in countries where compliance is not a concern.

Table 1 provides summary statistics of the firm variables used in the first part of the analysis. First,

the outcome variable is the (logged) real monthly wage bill (including allowances) plus other labour

costs (social and health insurance, training, recruitment costs etc.).13 In addition to the fact that non-

wage benefits constitute an important part of the total compensation package in Vietnam (Quang,

2008), the rationale for including other labour costs in the total wage, is that one would expect

unions to be positively associated with the provision of non-wage (fringe) benefits, as found in Rand

and Tarp (2011).14

Table 1: Summary statistics, firm characteristics

Using the monthly wage as opposed to hourly wages may lead to bias on the

union coefficient if unionization affects the number of hours worked. However, this issue is

addressed by using a number of control variables (described below). Moreover, in the second part of

the analysis, I use the individual wage which is calculated based on hours worked.

Panel A (unbalanced panel)

Total 2007 2009 Mean SD Mean SD Mean SD Total monthly real wage (logged) 9.421 1.146 9.212 1.163 9.635 1.090 Basic monthly real wage (logged) 9.372 1.126 9.167 1.140 9.582 1.076 Monthly real other labour costs (logged) 3.934 3.671 3.521 3.603 4.357 3.702 Trade Union 0.322 0.468 0.310 0.464 0.335 0.474 Firm size (regular full-time employees) 32.621 32.757 30.959 31.504 34.323 34.003 Private firm = 1 0.272 0.446 0.316 0.466 0.228 0.421 Cooperative/Collective/Partnership = 1 0.089 0.285 0.099 0.300 0.078 0.269 Limited liability company = 1 0.553 0.498 0.526 0.501 0.581 0.495 Joint stock company = 1 0.086 0.280 0.058 0.235 0.114 0.318 Owner is male 0.651 0.477 0.661 0.475 0.641 0.481 Owner has higher education 0.867 0.340 0.848 0.360 0.886 0.318 Professional share 0.084 0.077 0.081 0.081 0.088 0.072

13 It should be noted that since the wage data is from the end of the previous year to which the survey was carried out, the union variable could be subject to endogeneity with better performing firms establishing unions. In order to check for this, I used the 2005 survey to compare the wages between firms that had established unions between the 2005 and 2007 survey, and those that had unions in both years. The median wage for the two groups is almost identical (differs on the fourth decimal), while the mean wage is slightly higher for those that had established unions between the two years. Thus, reverse causality does not seem to be a major concern. 14 In a survey of the literature (mostly US based), Lewis (1986) finds that omissions of fringe benefits lead to a downward bias on the union wage gap of between 2 and 3 pct., while Freeman and Medoff (1984) show that there is a union fringe benefit differential analogous to the union wage gap.

13

Female share 0.392 0.246 0.377 0.242 0.408 0.249 Temporary worker share 0.111 0.185 0.108 0.191 0.115 0.179 Location (Urban = 1, Rural = 0) 0.574 0.495 0.585 0.494 0.563 0.498 Location (North = 1, South = 0) 0.521 0.500 0.526 0.501 0.515 0.501 Provision of social insurance = 1 0.500 0.501 0.439 0.498 0.563 0.498 Total number of observations 338 171 167

Panel B (balanced panel)

Total 2007 2009 Mean SD Mean SD Mean SD Total monthly real wage (logged) 9.510 1.048 9.378 1.019 9.641 1.065 Basic monthly real wage (logged) 9.456 1.026 9.328 0.994 9.585 1.047 Monthly real other labour costs (logged) 4.073 3.696 3.807 3.606 4.340 3.784 Trade Union 0.394 0.490 0.372 0.486 0.415 0.495 Firm size (regular full-time employees) 35.436 32.529 36.117 32.421 34.755 32.795 Private firm = 1 0.234 0.425 0.255 0.438 0.213 0.411 Cooperative/Collective/Partnership = 1 0.096 0.295 0.096 0.296 0.096 0.296 Limited liability company= 1 0.606 0.490 0.596 0.493 0.617 0.489 Joint stock company = 1 0.064 0.245 0.053 0.226 0.074 0.264 Owner is male 0.633 0.483 0.628 0.486 0.638 0.483 Owner has higher education 0.894 0.309 0.894 0.310 0.894 0.310 Professional share 0.084 0.075 0.079 0.076 0.089 0.075 Female share 0.390 0.237 0.382 0.235 0.398 0.240 Temporary worker share 0.118 0.186 0.122 0.188 0.115 0.184 Location (Urban = 1, Rural = 0) 0.553 0.498 0.553 0.500 0.553 0.500 Location (North = 1, South = 0) 0.495 0.501 0.495 0.503 0.495 0.503 Provision of social insurance = 1 0.500 0.501 0.500 0.503 0.500 0.503 Total number of observations 188 94 94 Note: 1 USD =

16,010 (31/12/2007) and 18.465 (31/12/2009). Real wages deflated using GDP deflators (base year=1994).

Table 1 shows that the (logged) monthly wage bill for a firm has increased substantially over time,

both for the unbalanced and balanced panel. Since the wage measure includes non-wage costs, one

potential explanation could be that the increase is attributed to higher social contributions following

the Law on Social Insurance (LSI) which came into effect on 1 January 2007.15 Table 1 indeed shows a

substantial increase in other labour costs between 2007 and 2009, yet as a share of wages there is

no substantial change.16 Rather, since basic wages have risen over time it seems that firms have

become more productive, perhaps as a result of hiring more highly skilled workers. Even if the LSI

has not led to increased non-wage costs, the fact that it requires social insurance

15 Recall from footnote 14 that the wage data for the 2007 survey is from end 2006.

contributions for

16 Moreover, the average firms’ expenditure on non-wage labour costs is around 6 pct. which is substantially below the 15 pct. social insurance contribution mandated by Law (Article 149). See Nguyen et al. (2006) for a comprehensive review of the Labour Code. Moreover, since non-wage costs include recruitment and training costs, the social security contribution part is presumably even lower.

14

full-time employees with contracts of more than 3 months (previously 12 months) might have

further increased the demand for better quality workers. In addition, the fact that minimum wages

determine actual wage policy at the enterprise level in Viet Nam might also contribute to higher

basic wages over time (there was a rise in the real minimum wage between 2007 and 2009). For the

balanced panel the wage is generally higher (as are other labour costs) suggesting higher

productivity for incumbent firms.17

Second, the main variable of interest is union status which is an indicator variable taking the value 1

if the firm has a union and 0 otherwise. I made use of having both firm and individual union

information in order to cross-check union status, so as to minimize any potential bias arising from

the misclassification of union status. 18

Table 1 Panel A shows that around 32 pct. of firms have

trade unions, which is lower than the 50 pct. (national) figure reported by the VGCL (VGCL, 2010),

yet given that I am working with small and medium firms the lower share is perhaps not surprising.

There has been a rise in the share of unionized firms between 2007 and 2009, by around 2.5

percentage points, which is also lower than the national figure of 5 percentage points, yet the latter

being for a longer time period from 2007 to mid-2010 (VGCL, 2010). In Panel B (the balanced panel)

the unionized share of firms is about 40 pct. (closer to the national level) indicating that incumbent

and more established firms are more likely to have unions. This may be related to the fact that firms

are only required to set up unions within 6 months of establishment (the unbalanced panel includes

new entrants). The increase over time is around 4 percentage points, which is closer to the national

figure of 5 percentage points.

I further control for firm size, legal status, the owners’ gender and education, the share of

professional workers, females and temporary workers, location and sector.19

17 An alternative way to measure wages would be to use wages per employee (average wages), yet given the enormous variation in wages between different worker types (see Larsen, Rand and Torm, forthcoming), the average wage would not have much practical meaning and therefore I decided to use the total wage bill.

The justifications for

the selection of these control variables and their summary statistics are as follows. First, in terms of

firm size, since this is measured by the number of full-time employees, the hours issue raised above

18 Inaccuracy of union status change may lead to bias in wage gap estimates (especially, in the case of panel estimations). Five firms indicated not having a union, yet all the workers represented for each firm reported being union members, thus these firms were assumed to have unions (note, this change did not alter the results significantly). Moreover, I used the 2005 data to check for firms that had indicated having a union both in 2005 and 2009 but not in 2007, and found 11 such cases. Removing these firms from the dataset did not alter the results reported in section 6, thus I kept them in. 19 Initially, I also included the union membership share in order to account for the fact that this may be associated with higher wages (union membership share rose from 73 pct. in 2007 to 81 pct. in 2009), yet the variable was insignificant and its inclusion did not alter the results.

15

is less concerning. Moreover, a general finding is that earnings tend to rise with firm size (Oi and

Idson, 1999; Söderbom et al., 2005), since larger firms have a productivity advantage due to scale

efficiency (Tybout, 2000). In addition, by controlling for firm size I am taking account of the fact that

the union markup may represent some form of compensating wage differential associated with a

possibly less salubrious work environment in larger firms. The average firm has 33 employees and

the size of firms has increased in between the two years. For the balanced panel the average size is

35, yet incumbent firms have shrunk in size. Combined with the increase in the wage bill it seems

that these firms might have shifted to employing fewer but more qualified workers.

Second, since there is substantial variation in wages and the degree of unionization across

ownership types, I incorporate the different legal categories. Moreover, ownership form has been

shown to be a critical factor influencing the adoption of human resource practices, including in

relation to trade unions (Zhu et al., 2008). Limited liability companies constitute the largest category

with around 55 pct. (60 pct. for the balanced panel), and the share has risen over time as has joint

stock companies (the smallest legal category). By contrast the share of both private firms and

cooperatives/collective/partnerships has fallen. Third, the gender of the owner has been shown to

be important in terms of compensation with female owners generally being more generous in the

provision of non-wage benefits (Rand and Tarp, 2011), I therefore include a dummy taking the value

1 if the owner is male and 0 if female. 65 pct. of the firms have male owners, but the share of

females owners is increasing over time, except for in the balanced panel where one firm has

changed to having a male owner between 2007 and 2009.

Fourth, studies have found that well-educated managers are more likely to hire well-educated

workers (Rosenbaum et al., 1999), and to capture this I include a dummy indicating whether the

owner has at least high school education. Table 1 shows that around 87 pct. of owners have higher

education (slightly higher for the balanced panel), and this ratio has risen over time, indicating a

general increase in the skill level of managers in accordance with increased productivity. Fifth, since

the average educational level in the firm has been shown to be positively correlated with overall

productivity and wages (Lucas, 1988), I include the share of professional workers (of the regular

workforce) as a proxy for the general quality of the workforce. Moreover, this eliminates the bias on

the union variable that might arise from a positive correlation between union status and worker

quality if, as a result of unionization firms, choose to hire better quality workers so that productivity

matches the union-imposed higher wage (Lewis, 1986).20

20 Pencavel (2005) summarizes a number of studies, where the union wage gap drops substantially once worker skill is controlled for.

Similarly, higher wages may attract more

16

highly qualified workers. Furthermore, given that workers with higher unobserved ability will tend to

have coworkers with higher average skills, including a measure for the latter will also help to reduce

the bias arising from omitted worker specific ability. Finally, as noted by Pencavel (2005), in a union-

setting, where wages are typically determined by some procedure for a reference worker or a

reference job wages may not be exactly related to all observable characteristics of any one worker,

but rather to broader attributes of a type of worker, as captured by firm specific worker shares.

Professionals make up around 8 pct. of the workforce and the ratio has risen over time, especially

for incumbent firms, providing support for the above argument that firms are hiring more skilled

labour (perhaps in part related to the requirements of the LSI), which in turns helps to explain the

increase in wages over time.21

Sixth, the share of female workers (of the regular workforce) is included since this has been found to

have a negative effect on the wages of all workers in the firm (Croson and Gneezy, 2009). The

summary statistics show that the proportion of females is just under 40 pct., and the share has risen

over time, in line with the above trend on gender of the owner (for the balanced panel the increase

in slightly lower corresponding to the fact that one firm has shifted to male ownership over time).

Seventh, temporary workers as a share of the total labour force is incorporated so as to account for

the impact this may have on the wage structure, since these workers may work shorter hours, and if

they are employed for less than 3 months they will not receive social insurance. Further, unionism

has been shown to be associated with a higher share of non-standard workers (Park, 1991) as firms

substitute them for more expensive unionized labour. The proportion of temporary workers is about

11 pct. and for the unbalanced panel this has increased slightly over time, perhaps related to the fact

that employing workers (for more than 3 months) has become more expensive with the Social

Insurance Law. By contrast, for the balanced panel, the share has fallen over time, which could

simply be an indication that as firms become more established they tend to rely increasingly on a

more permanent/regular workforce.

Eight, the inclusion of location as a control variable is important in order to account for the fact that

Vietnamese provinces are relatively autonomous, with substantial institutional differences. This

aspect is also well documented in the Provincial Competitiveness Index (Malesky, 2009). I model

21 Although at this stage there is no evidence that professionals earn higher wages than other worker types, this will become clear once the employee data is introduced in the results section. I also added the shares of other worker categories (managers, sales, office, service and production workers), yet none of there were significant and their inclusion did not alter the results.

17

location using 10 indicator variables representing whether the firm is in a given province. Table 1

shows that around 57 pct. of firms are located in urban areas, and 52 pct. in Northern provinces

(both lower for the balanced panel). Ninth, I also include sector dummies (based on 2-digit level ISIC

codes) to control for differences in general conditions among the different sectors. Moreover, union

wage differentials are likely to vary across sectors due to differences in economic rents between

product markets, and variations in the surplus available to firms. Sector summary statistics are

provided in Appendix Table A.22

Table 2 provides summary statistics of the employee variables used in the second part of the analysis

(the corresponding summary statistics of the firm characteristics are provided in Appendix Table B).

Finally, (only) half of the firms contribute to social insurance for

their employees, yet for the unbalanced panel the share has risen by more than 12 percentage

points over time, perhaps an indication that the LSI is taking effect, at least for new firm entrants.

Table 2: Summary statistics, employee characteristics

Panel A (unbalanced panel)

Total 2007 2009 Mean SD Mean SD Mean SD Monthly real wage (logged) 6.433 0.435 6.398 0.442 6.458 0.429 Social Insurance 0.529 0.499 0.470 0.500 0.570 0.496 Member of trade union 0.351 0.478 0.304 0.460 0.385 0.487 Member of trade union if firm has a

0.851 0.357 0.775 0.418 0.900 0.301

Gender (male = 1) 0.524 0.500 0.516 0.500 0.530 0.500 Age 33.71 9.844 34.151 10.382 33.410 9.442 Higher education 0.785 0.411 0.799 0.401 0.776 0.418 Manager 0.140 0.347 0.149 0.356 0.133 0.340 Professional worker 0.169 0.375 0.195 0.397 0.151 0.358 Sales worker 0.108 0.311 0.130 0.337 0.093 0.291 Production worker 0.375 0.484 0.298 0.458 0.430 0.495 Other (office and service workers) 0.208 0.406 0.229 0.420 0.194 0.396 Informal 0.640 0.480 0.539 0.499 0.712 0.453 Total number of observations 1153 477 676

Panel B (balanced panel)

Total 2007 2009 Mean SD Mean SD Mean SD Monthly real wage (logged) 6.515 0.441 6.509 0.455 6.522 0.428 Social Insurance 0.664 0.473 0.664 0.475 0.664 0.475 Member of trade union 0.602 0.491 0.513 0.502 0.690 0.464 Member of trade union if firms has a

0.877 0.329 0.806 0.399 0.940 0.239

Gender (male = 1) 0.522 0.501 0.522 0.502 0.522 0.502 Age 35.37 9.85 34.28 9.82 36.46 9.80 Higher education 0.841 0.367 0.841 0.368 0.841 0.368 Manager 0.181 0.386 0.186 0.391 0.177 0.383 Professional worker 0.217 0.413 0.212 0.411 0.221 0.417

22 The largest sectors represented in the sample are food processing (ISIC 15) (16 per cent), fabricated metal products (ISIC 28) (14 per cent) and rubber (ISIC 11) (11 per cent). For presentational purposes I do not report specific sector level results, but these are available upon request.

18

Sales worker 0.155 0.363 0.142 0.350 0.168 0.376 Production worker 0.252 0.435 0.248 0.434 0.257 0.439 Other (office and service workers) 0.195 0.397 0.212 0.411 0.177 0.383 Informal 0.668 0.472 0.575 0.497 0.761 0.428 Total number of observations 226 113 113

Note: 1 USD =

First, the outcome variable is the (logged) monthly equivalent of the real individual basic wage

(excluding social benefits) calculated based on an 8 hour work day and 6 working days a week (the

sample median), thus avoiding any bias arising if union workers work longer (or shorter) hours than

non-union members.

16,010 (31/12/2007) and 18.465 (31/12/2009). Real wages deflated using GDP deflators (base year=1994).

23 Moreover, since the employee survey asked for the current wage (in the firm

case the wage data were from the end of the previous year) this gets around any timing issue. Table

2 Panel A shows that the (logged) monthly real wage has risen over time, yet the magnitude of the

increase is much smaller than for firm wages in Table 1, and the reasons for this will become clearer

below.24

For the balanced panel (Panel B), the average wage is higher implying that tenure is

positively related to wages, as would be expected. The second outcome variable to be used in the

analysis (in section 6) is social insurance, modeled as an indicator variable taking the value 1 if the

worker in question receives social insurance and 0 otherwise. For the unbalanced panel 53 pct. of

workers report receiving social benefits, and the proportion has risen over time consistent with the

firm level results in Table 1. In the balanced panel (Panel B), the share of workers provided with

social benefits is substantially higher (66 pct.), indicating that fringe benefits, like wages, are

associated with time spent in firm.

Third, my variable of interest, union membership, takes the value 1 if the worker is member of a

union and 0 otherwise.25

23 In fact, I found no evidence that union members differed from non-union members in terms of hours worked (in the sample both union and non-union workers work on average 8 hours per day).

Table 2 shows that union membership has increased over time, especially

for the balanced panel, for which union membership is also substantially higher compared with the

unbalanced panel, consistent with other studies showing that union membership is related to tenure

(see for instance Card and de La Rica, 2006). Also, the higher level of unionization in the balanced

panel it is an indication that workers that were interviewed in both 2007 and 2009 have been

sampled from more established firms with higher levels of unionization. These trends match the

national trend of an increase in union membership over this period (VGCL, 2010). Within unionized

24 When converting to the normal scale (figures not reported), it is reassuring to note that (for the unbalanced panel) the average wage for an individual worker (685 thousand VND per month) is similar to the average firm wage in Table 1 (22884.7 thousand VND divided by the average firm size of 32.6). 25 In order to minimize the risk of measurement error, I scrutinized (individual) observations indicating union membership, but where the corresponding firm did not have a union. As mentioned in an earlier footnote there were five firms which indicated not having a union, while several workers in each firm reported being union members, thus these firms were assumed to have unions (this did not alter the results significantly).

19

firms membership is extremely high, and has increased over time, suggesting that there must be

certain gains associated with being a union member.

Fourth, given that it is common to find gender wage gaps, in particular in developing countries

(Jones, 2001), I incorporate a gender dummy (equal to 1 if the worker is male). Table 2 shows that

the gender division is close to equal and women constitute a larger share of the employee sample

than the firm sample. Fifth, I control for age of the worker as a proxy for experience – a key variable

in the standards human capital earnings function (Mincer 1974)26

Sixth, I include the workers’ education level (a dummy which equals 1 if the individual has at least a

high school/college degree, and 0 otherwise) since according to both human capital (Mincer, 1974)

and signaling theory (Spence, 1973) the level of education accounts for a large share of the variation

in earnings. Moreover, as noted earlier unionized firms may hire better quality workers. The share of

workers with a higher education is around 80 pct., and even more for the panel of workers.

, and include age squared to allow

for a diminishing marginal effect. The average age is very similar for the two panels, and reassuringly

for the balanced panel the average age has increased by almost exactly two years in between the

two survey years. The slight discrepancy (at the decimal level) may be attributed to the fact that not

all workers were interviewed on exactly the same date in both years, (thus some workers would

have aged by more than 2 years for the 2009 survey). Given this minor divergence seasonal factors

(which could confound the analysis) will be negligible.

Seventh, the different job functions are included as dummy variables on the basis that wages have

been shown (Larsen, Rand and Torm, forthcoming) to vary substantially between occupation

categories (beyond what is captured by controlling for education). Panel A in Table 2 shows that

managers account for around 14 pct., professional for 17 pct., sales workers for 11 pct., production

workers for 38 pct. and others (office and service workers) for 21 pct. In the balanced panel, the

share of the first three worker types are higher, whereas the latter two are lower, particularly for

production workers suggesting that job function is related to tenure. 27

26 The question on years of experience was only added in the 2009 survey, and the measure of tenure unfortunately suffers from severe measurement error (58 workers have indicated a tenure that does not match across the two periods) and therefore I cannot use the variable.

For the unbalanced panel,

the shares of all worker types, except production workers, have fallen from 2007 to 2009. In the case

of professional workers this is in contrast to the rising share shown in Table 1, and such differences

27The interviewed workers were selected randomly, but the employee sample is not perfectly representative in terms of occupation categories. However, controlling for job functions in the empirical analysis helps to eliminate the bias that may arise from the sample not being representative.

20

in worker type trends between the firm and the employee samples help to explain why the wage

increase over time in the latter sample is substantially smaller than for the firm sample. I shall return

to this when discussing the results in section 6.

Finally, I include a dummy variable for whether the worker found their job through an informal

contact (knowing the owner or someone who works in the firm), as opposed to via a formal contact

(advertisement, employment agency etc.), since this has been shown to be associated with higher

individual wages (Larsen, Rand and Torm, forthcoming). 64 pct. of workers have been hired through

an informal contact, and this method of finding a job has increased over time. Interestingly, the

share has also increased for the balanced panel (presumably the same worker was only hired once),

yet this could be related to the fact that one of the informal job finding categories changed in the

2009 survey to include acquaintances as well, and it is precisely in this category

(“friend/acquaintance working in the firm”) that the largest increase has been (from 20 to 38 pct.),

whereas there has been an equivalent drop in those replying “door to door visits”. Thus, if those that

were hired through door to door visits were in fact hired through acquaintances (and indicated the

latter in the 2009 round of the survey), this would explain the observed trend.

As for the firm characteristics corresponding to the worker sample, all the variables as described in

connection with Table 1 are included, and their summary statistics (for the unbalanced panel) are

provided in Appendix Table B. Briefly, in terms of firm size, workers have been sampled mostly from

larger firms with on average 40 employees (compared with 33 for the firms in Table 1). With regard

to the legal categories, the workers have been sampled from firms in a manner that is relatively

consistent with the firm distribution. Moreover, employees have been sampled mostly from female

headed firms (as revealed also by the higher share of females), and firms where the owner has a

higher education. The share of professionals has fallen over time, which is in accordance with the

results in Table 2 Panel A, yet as mentioned earlier this is in contrast to the firm trend (Table 1).

Finally, workers have been surveyed more frequently in urban firms.28

6. Results

6.1. Firm wages

28 Appendix Table C presents the union membership probit estimations underlying the analysis in section 6. The results show that differences between union members clearly exist along the following dimensions: age of worker, firm size, owner’s education, legal category 3 (cooperatives, collectives, partnerships) and 4 (Limited liability company), location (south), and sector (not reported).

21

First, I consider the firm union wage association (specification (i) in section 5) for the unbalanced

panel. Recall that the wage measure is the total wage bill (including allowances) plus other labour

costs (social and health insurance, training, recruitment costs etc.). Table 3 shows that firms that

have trade unions have average wages that are around 18 pct. higher when all control variables are

included (column 2). Since there are no sector level bargaining agreements, the wage premium is

relative to a non-union situation29

.

In terms of the control variables, firm size is significantly positive in line with the general finding that

larger firms pay higher wages (Oi and Idson, 1999; Söderbom et al., 2005). Among the legal

categories CCP (cooperatives/collectives and partnerships) have significantly lower wages compared

with private firms (the reference category). As for the different worker shares, the share of

professional workers has the expected positive sign, and the coefficient indicates that going from 0

pct. to 100 pct. professional workers is associated with a wage bill that is 75 pct. higher. By contrast

the share of female workers has the expected negative sign, yet is not well-determined. As for the

temporary worker share, this is associated with higher average wage, as expected.30 The year

dummy indicates that wages have risen by about 28 pct. between 2007 and 2009 which, as

discussed earlier, is an indication of increasing productivity and firms employing more highly skilled

workers.31

Table 4 presents the results for the balanced panel of firms. The first column shows the OLS results

with all controls included, and here the trade union coefficient is 8 percentage points higher

compared with in the unbalanced sample. In terms of the other variables (not reported), they are all

as expected and in line with the results in Table 3. Column 2 and 3 report the fixed effects (FE)

estimates, controlling for the possibility that unobserved firm specific heterogeneity is driving the

Moreover, as highlighted earlier, the fact that the wages of all types of workers are fi xed

through a multiplier of the minimum wage (which rose by 8 pct. in real terms between 2007 and

2009) could also have contributed to the real wage increase. Furthermore, the observed wage

increase is comparable to China, a country with which Vietnam shares many features, and where

real wages grew by an average of 13 pct. per annum between 2001 and 2007 (ILO, 2008).

29 In cases where the alternative is a sectoral bargaining agreement, this would be the benchmark. 30 By definition a higher temporary worker share implies a lower regular workforce share (since these two components make up the total labour force), which in turn leads to a higher average wage per regular worker (lower denominator). Thus, temporary workers do not actually get paid more, quite the opposite (the correlation between the temporary worker share and the total wage bill is negative). 31 Note that when the full sample of firms is considered including those where employees have not been sampled, real wages have “only” risen by 20 pct. between 2007 and 2009, which is in line with the manufacturing wage figures from the GSO enterprise data (2000 to 2006) (Phan, 2009).

22

OLS estimates. In the FE estimation with all controls, the union membership coefficient is about 3.5

percentage points higher, indicating that the OLS estimates are downward biased, and implying that

the unobserved heterogeneity is negatively correlated with unionization. A possible explanation for

this finding is that those firms that are less likely to be unionized are also relatively poor performers

(alternatively unionized firms are more likely to be the better performing ones). I will return to this

issue in sub-section 6.2 on the individual union membership-wage relation. All in all, the OLS and FE

estimates reveal a positive association between firm unionization and wages.

Table 3: Firm wages (unbalanced panel)

(1) (2) OLS OLS Trade union 0.172** 0.183** (0.081) (0.072) Firm size (log employment) 1.024*** 1.035*** (0.039) (0.041) CCP -0.498*** (0.125) Limited liability 0.006 (0.086) Joint stock company 0.036 (0.109) Owner male 0.043 (0.061) Owner has higher education 0.110 (0.085) Professional worker share 0.754** (0.350) Female worker share -0.199 (0.165) Temporary worker share 0.920*** (0.214) Year dummy 0.293*** 0.278*** (0.059) (0.056) Province dummies No Yes Sector dummies No Yes R-squared 0.776 0.845 Observations 338 338 Note: Dependent variable: Log real wages. OLS. Robust standard errors (in parenthesis). *** p<0.01, ** p<0.05, * p<0.1

Table 4: Firm wages (balanced panel)

(1) (2) (3) OLS FE FE Trade union 0.265*** 0.319** 0.300* (0.099) (0.153) (0.156) Firm size 0.942*** 0.791*** 0.782*** (0.059) (0.100) (0.111) Owner male Yes No Yes Owner education Yes No Yes Legal dummies Yes No Yes Worker shares Yes No Yes

23

Province dummies Yes No No Sector dummies Yes No No R-squared 0.839 0.567 0.605 Observations 188 188 188 Note: Dependent variable: Log real wages. OLS and FE estimates. Year dummy included. Robust standard errors (in parenthesis). *** p<0.01, ** p<0.05, * p<0.1

As discussed in the previous section, the Law on Social Insurance (2007) has meant that firms are

more likely to incur higher non-wage costs, thus it is possible that the higher compensation

associated with unionization, as shown in Table 3 and 4, is driven by higher non-wage benefits,

rather than a higher basic wage (since the wage measure used includes other labour costs). When

using the basic wage (net of benefits) as the outcome variable (results are presented in Appendix

Table D) the union wage gap for the unbalanced panel is significant only at the 10 pct. level and only

when all controls are included. For the balanced panel, however, the union wage association is

significant at the 5 pct. level, suggesting that the relation is stronger in more established firms.32 In

sum, however, the difference between union and non-union firms seems to be particularly robust

when the provision of non-wage benefits is taken into account. This is further supported by the

results presented in Table 5 showing a positive and highly significant association between

unionization and (logged) other labour costs, for both the unbalanced and balanced panel (columns

1 and 2 respectively).33

Table 5: Non-wage labour costs

(1) (2) Tobit Tobit Trade union 4.081*** 4.712*** (0.607) (0.769) Basic wages 2.444*** 2.664*** (0.526) (0.719) Firm size Yes Yes Owner male Yes Yes Owner education Yes Yes Legal dummies Yes Yes Worker shares Yes Yes Province dummies Yes Yes Sector dummies Yes Yes Observations 338 188 Note: Dependent variable: Log other labour costs. Tobit. 151 and 81 left censored variables respectively in columns 1 and 2. Year dummy

32 The (basic) wage coefficients in Table D are about 5 percentage points lower than the wage coefficients reported in Tables 3 and 4, which is consistent with the downward bias findings summarized in Lewis (1986). I further note that when considering the full firm sample (including firm from which employees were not sampled), the union basic wage premium is insignificant in all specifications, indicating that the subsample of firms from where employees have been surveyed is not representative of the full firm sample. 33 Recall that other labour costs in addition to social and health insurance also include training and recruitment costs, thus the positive association could also be related to unionized firms providing more training or having higher hiring costs.

24

included. Robust standard errors (in parenthesis). *** p<0.01, ** p<0.05, * p<0.1

Table 5 also shows a positive correlation between basic wages and other labour costs, which is

logical given that social contributions are calculated as a percentage of the basic wage (15 pct. in

theory, but substantially less in practice as revealed in Table 1). In terms of the other variables (not

reported), education of the owner, the share of professionals and the share of females are all

positively associated with non-wage costs. That unions serve to protect workers through ensuring

the provision of social benefits is a positive finding, and reflects their main function under central

planning, namely being responsible for the social welfare of the workers. However, as seen in Table

1 only around 50 pct. of firms actually contribute to social insurance (firms with more than 10

employees are mandated by Law to contribute to social insurance), and although this share has risen

over time, perhaps in part due to the Law on Social Insurance, non-compliance is prevalent.34

Another measure of union power is the wage share (Boeri and van Ours, 2008). However, I find no

significant relation along this dimension, whether using total or basic wages as the outcome

measure (results based on the total wage measure are reported in the Appendix Table E), indicating

that unionization does not enable workers to obtain a “larger share of the pie”. This is perhaps not

surprising since, as in other countries, Vietnam’s openness to increasing international competition

means that unions are less able to extract rent from firm profits (Benson

and Gospel, 2008).

Moreover, Pencavel (2005:79) states that generally wage increases are not enjoyed at the expense

of firm profits since in most economies profits represent only a small fraction of total income, and if

these are distributed over the entire workforce, they constitute only a small addition to wages.

Pencavel further argues that firms generally do not passively respond to attempts to appropriate

their surpluses, but would aim to recover any loss of profits through union activity through higher

prices or lower employment.

Since the firm wage bill measure is an aggregation of individuals wages, the internal wage

distribution is concealed, and thus in the next sub-section I make use of the information on

individual union membership and wages, as well as more detailed employee specific information in

order to examine the union wage gap further.

34 This is in accordance with Nguyen et al. (2006) who note that in Vietnam 56 pct. of the eligible workforce is covered by social insurance and that nonpayment is particularly prevalent among domestic private firms. This is also supported by the figures in Table 2 showing that 53 pct. of workers (unbalanced panel) receive social insurance (substantially higher for the balanced panel).

25

6.2. Individual wages

As pointed out in section 5, since the employee sample consists of workers in both union and non-

union firms, selection bias may arise both due to the fact that workers self-select into a firm that has

a union, and given that the firm has a union they self-select into union membership – if they believe

this to be beneficial. However, as noted earlier, in the employee questionnaire when asked about

the main benefit of being a union member only 9 pct. answered “better and more stable wages”

suggesting that workers do not generally select into unions in order to get higher wages.35

First, I examine the union wage differential for the unbalanced worker sample where the outcome

variable is the individual wage and the variable of interest is union membership. Table 6 presents the

results revealing a significant individual wage premium associated with being a union member: 12

pct. when only employee characteristics are included (column 1) and just under 7 pct. when firm

characteristics are added (column 2). The real wage gain could be slightly lower, since presumably

the union membership contribution, which amounts to 2 pct. of wages, has not been deducted in

the reported wage. However, the results are largely in accordance with Clarke et al. (2007), who

report that Vietnamese trade unions were able to negotiate an increase in workers’ wages of 5 pct.

36

Since the wage measure used here is the basic wage, the premium is not driven by other labour

costs (social benefits, training etc.). This may also explain why the union coefficients are more in line

with those from the basic wage firm level results (Appendix Table D).

In terms of the control variables, there is a substantial gender wage gap with male earnings being

about 15 pct. higher than for females, which is a common finding in earnings analysis (Jones, 2001)

and corresponds to other Vietnamese studies (Liu, 2004). The age of the worker is also highly

significant and has the expected concave effect with a maximum at around 43 years of age. With

35 The fact that wages are not ranked as the most valuable aspect of unions activities has also been found in other studies (Pencavel, 2005: 85-86) suggesting that as countries grow richer the focus may shift to other aspects of the employment relationship, including fringe benefits and safety issues. In order to rule out selection bias a Heckman two-step procedure was carried out revealing no evidence of sample selection. Results are available upon request. 36 Similarly, in the case of Spain, Card and De La Rica (2006) showed that unions were associated with a 5-10 pct. individual wage premium.

26

regard to education, the results show that a worker with education beyond secondary school has a

wage that is about 12 pct. higher than a worker with only basic education. Thus, the characteristics

of the workers are all in line with human capital (Mincer, 1974) and signalling theory (Spence, 1973)

Table 6: Individual wages

.

(1) (2) OLS OLS Union member 0.121*** 0.067*** (0.023) (0.025) Male 0.177*** 0.151*** (0.023) (0.022) Age 0.031*** 0.035*** (0.008) (0.007) Age squared -0.039*** -0.041*** (0.010) (0.009) Higher education 0.138*** 0.120*** (0.028) (0.027) Manager 0.508*** 0.465*** (0.041) (0.038) Professional 0.306*** 0.276*** (0.035) (0.031) Sales 0.207*** 0.171*** (0.042) (0.039) Other 0.109*** 0.052* (0.030) (0.029) Informal contact 0.094*** 0.107*** (0.023) (0.022) Firm size (log employment) 0.086*** (0.015) CCP -0.147*** (0.048) Limited liability -0.012 (0.030) Joint stock company 0.028 (0.042) Owner is male 0.004 (0.022) Owner has higher education 0.095** (0.045) Professional worker share 0.168 (0.150) Female worker share -0.204*** (0.065) Temporary worker share 0.037 (0.058) Year dummy 0.064*** 0.055** (0.023) (0.022) Province dummies No Yes Sector dummies No Yes R-squared 0.295 0.429 Observations 1153 1153 Note: Dependent variable: Log real individual wages. OLS. Robust standard errors (in parenthesis). *** p<0.01, ** p<0.05, * p<0.1

The different occupation categories all indicate a substantial wage premium compared to production

27

workers (the reference group), especially for managers and professional workers at close to 50 and

30 pct. respectively. Having been hired through an informal contact, gives a positive wage return of

about 10 pct. compared with a formal method, which is in line with findings in Larsen, Rand and

Torm (forthcoming). With regard to the firm variables, these are all in accordance with Table 3. Firm

size is significantly positive and individuals working in a CCP have significantly lower wages

compared with workers in private firms. The gender of the firm owner is not well-determined, yet a

more highly educated owner is positively associated with wages. The share of professional workers

is insignificant (but positive), possibly due to being accounted for by the individual education

variables.37 The share of females is significantly negative which is in line with similar studies (Card

and De La Rica, 2006). The temporary worker share is insignificant, which is related to the fact that

unlike in the firm case, in the worker sample this variable is not directly related to the way the wage

measure is constructed. 38

Further, the year dummy shows an increase in individual (real) wages of about 6 pct. between 2007

and 2009, which is substantially below the 28 pct. revealed for the firm level results in Table 3. In

other words, and as already discussed in relation to the summary statistics, individual wages (for the

average worker in this sample) have not risen as much over time as total firm wages. This

discrepancy arises from differing trends in the workforce structure between the firm and employee

samples combined with the substantial variation in wages between different worker types. The

summary statistics in Table 2 revealed that professionals, who receive a significantly higher wage

than production workers (Table 6), have seen a declining share between 2007 and 2009 (also as

shown in Appendix Table B employees have been sampled mostly in firms with a declining share of

professionals), yet at the firm level (Table 1) the professional share has risen39

37 Controlling for a worker’s observed skills, a higher/lower level of coworker skills implies that the worker has above average/below average unobserved skill characteristics, and thus removes some of the potential bias arising from omitted unobserved factors.

. As for production

workers, who receive a lower wage compared with the other worker categories (Table 6), their share

in the employee sample has risen by about 13 percentage points compared with only 2 percentage

points in the firm sample (and their wages – not reported – have not risen to the same extent as the

other worker types). Essentially, since the employee sample is not representative of the firm sample

38 However, it remains positive which is reassuring since if the wage share of real value added is fixed, individuals employed in firms with a higher share of temporary workers are likely to earn more. 39 The professional workers represented in the employee sample have seen a rise in real wages of 20 pct. (not reported). Moreover, managers and sales workers, who also earn relatively higher wages (Table 6), have seen a decline in the unbalanced employee sample (Table 2) to a much larger extent than in the firm sample (not reported).

28

along the occupation dimension it is unable to capture firm wage differences resulting from changes

in the workforce structure.40

Had all employees participated in the employee survey the trends in

workforce compositions would presumably be comparable across the samples and the results similar

over time.

Table 7 presents the OLS and FE results based on the balanced panel of workers, including the same

employee and firm characteristics as in Table 6. Column 1 shows that the union wage gap is around 4

pct. points higher than for the unbalanced sample (column 2 in Table 6) suggesting larger wage

disparity among this smaller subsample of workers. The various employee and firm characteristics

are all as expected (not reported) and in line with the results in Table 6.

Table 7: Individual wages

(1) (2) (2) OLS FE FE Union member 0.109** 0.133* 0.153** (0.051) (0.070) (0.068) Employee characteristics Yes Yes Yes Firm characteristics Yes No Yes Province dummies Yes No No Sector dummies Yes No No Observations 226 226 226 R-squared 0.681 0.163 0.248 Note: Dependent variable: Log real individual wages. OLS and FE. Year dummy included. Robust standard errors (in parenthesis) clustered by worker id. *** p<0.01, ** p<0.05, * p<0.1

Aside from the potential bias from selection into union status (which was controlled for using the

Heckman procedure) the OLS results may be biased due to unobserved worker or firm specific

heterogeneity (for instance worker ability/motivation) being correlated with union status (beyond

that which is captured by controlling for individual observed quality and average worker quality).

Columns 2 and 3 in Table 7 report the fixed effects estimates, controlling for the possibility that any

such relations may be driving the results. In terms of the employee characteristics, the specifications

include a control for the worker having higher education (since some individuals have upgraded their

education in between the two surveys) and the occupation categories (as some workers have

changed job functions over time). With regard to the firm variables, in addition to firm size these

include the various worker shares and legal dummies since there are variations over time along

these dimensions. In the fixed effect estimation, the union membership coefficient is higher by

40 Moreover, firm wages are from end-2006 and end-2008, whereas individual wages are from mid-2007 and mid-2009, which might also lead to some differences in wage outcomes.

29

about 4 pct. compared with the OLS estimate when all controls are included, which is in line with the

firm results reported earlier, again indicating that the OLS estimates are downward biased, and in

turn implying that the unobserved heterogeneity is negatively correlated with union status. At the

firm level the explanation was that firms which are less likely to be unionized are also relatively

poorer performers, and here the corresponding interpretation is that lower ability workers are less

likely to be union members.41

The above analysis compares the wages between union members (in union firms) and non-union

members (in union and non-union firms). However, non-union members working in non-union firms

and union members in union firms may not be directly comparable, since the former have not been

faced with the choice of becoming union members (unless they purposely chose to work in a non-

union firms). Thus, this may not be a true indication of the individual wage gain from union

membership. In other words, the results do not reveal whether the wage differential is due to the

gap between workers in union and non-union firms, or the gap between union and non-union

workers within unionized firms. However, by making use of the information on firm unionization, I

am able to test this by restricting the sample to firms which have a union, making use of the fact that

not all workers in unionized firms are union members.

42

The results (reported in Appendix Table F) show that for the unbalanced panel the union member

variable within unionized firms is not well-determined, yet remains significant at a 10 pct. level for

the balanced panel. Thus, although there is some evidence of a reallocation of resources between

those workers that are union members and those that are not, this is only for workers who remained

41 This is supported by union membership being positively correlated with higher education, and lower skilled workers being relatively more likely not to be union members. In fact, when splitting the sample into production versus non-production workers, the union wage premium is significant only among non-production workers including all controls. Results available upon request. The fact that wages are raised more for higher level workers has also been found in other studies (Card and De La Rica, 2006; Dell’Aringa and Lucifora, 1994). 42 Recall from Table 2 that about 85 pct. of workers in unionized firms are union members, thus the amount of variation in the union membership variable is unfortunately rather limited. An additional scenario which could be examined with the matched employer-employee panel is the case where the firm changes union status over time, but the workers do not. However, there are only three cases where the firm changes from union to non-union status and the workers are not union members in either year. In terms of firms becoming unionized, there is only one case where the workers do not change union status. Thus, for the panel of workers, there is at the firm level extremely limited variation in union status between years; these firms either have or do not have unions and this is generally constant over time. The opposite scenario where the firm remains non-unionized, yet the workers change status is not realistic, since it is not possible to be a union-member if the firm does not have a union.

30

employed in both surveys, suggesting that union membership may pay off with time.43 However, all

in all, the results for whether there is a difference in wages between workers that are union

members and non-union members within unionized firms are not robust.44

The absence of a statistically robust wage difference between union-members and non-members in

unionized firms seems to suggest that these firms are characterized by a compressed wage

structure. What this implies is an extension of the above statement (following Table 7): lower ability

workers are less likely to be union members, because they are sorted into non-union firms (which

are characterized by lower performance), or alternatively higher ability workers tend to work in

unionized firms (which perform better and provide higher wages and social benefits).45

If non-union

members, within unionized firms, are able to obtain similar wages as union members then why do

most workers decide to become union members? Explanations of this union free-rider puzzle have

included the exclusive provision to union members of private goods (so-called “excludable incentive

goods”) such as on-the-job-training (Acemoglu, Aghion and Violante, 2001), legal and pensions

advice (Booth and Chatterji, 1995) or unemployment benefits (Brugiavini et al., 2001).

Recalling the results from the firm level analysis in section 6.1, which showed that unionization is

associated with higher non-wage benefits, and the fact that in the employee survey (both 2007 and

2009) 50 pct. of workers replied that the main benefits of trade union membership is that “it secures

that the firm pays social benefits”, it seems highly plausible that individual union membership is

positively associated with receiving non-wage benefits. Table 8 shows the Probit estimates where

social insurance is the dependent variable. The results show that being a union member is highly

positively correlated with receiving social benefits when controlling for both employee (column 1)

and firm characteristics (column 2). In terms of the other variables more highly educated workers,

managers and professionals have a higher probability of receiving benefits (they are also more likely

to be union members). Moreover, individuals working in larger and Southern firms, CCPs, joint stock

companies and in firms with a higher share of professional workers all have a higher probability of

43 I cannot exclude the possibility that a larger sample may reveal a statistically significant difference for the unbalanced sample. When restricting the sample further to only include those individual that became union members in between 2007 and 2009 (50 observations), the union coefficient remains insignificant. 44 In the case of South Korea Park (1991) finds that non-union members are paid less than union members, within unionized firms. However, in his study non-union members are mostly non-regular workers, whereas in my employee sample they are all regular workers. 45 This is further supported by the fact that the share of workers with a higher education is 0.81/0.88 in unionized firms compared with 0.77/0.75 in non-unionized firms for the unbalanced/balanced panels respectively.

31

receiving benefits. Finally, a higher share of female workers is positively related to the provision of

social benefits, as also found in Rand and Tarp (forthcoming), and in accordance with the negative,

yet insignificant, coefficient on owner being male46

.

Table 8

(1) (2) Probit Probit Union member 0.523*** 0.594*** (0.025) (0.029) Male -0.006 0.012 (0.035) (0.040) Age 0.011 0.018 (0.011) (0.012) Age squared -0.015 -0.022 (0.014) (0.016) Higher education 0.084* 0.090* (0.043) (0.049) Manager 0.152*** 0.128** (0.050) (0.057) Professional 0.109** 0.117** (0.050) (0.055) Sales 0.065 0.021 (0.056) (0.064) Other 0.088* 0.034 (0.047) (0.054) Informal contact -0.004 0.070* (0.035) (0.039) Firm size (log employment) 0.089*** (0.026) CCP 0.400*** (0.037) Limited liability -0.131 (0.086) Joint stock company 0.284*** (0.051) South 0.293*** (0.052) Owner is male -0.017 (0.038) Owner has higher education 0.057 (0.067) Professional worker share 0.905*** (0.264) Female worker share 0.308*** (0.098) Temporary worker share -0.114 (0.109) Sector dummies No Yes Observations 1,149 1,149 Note: Dependent variable: worker receives social insurance. Probit estimates, marginal effects. Year dummy included. Robust standard errors (in parenthesis) .*** p<0.01, ** p<0.05, * p<0.1

The results in Table 8 compare union members (in union firms) and non-union members (in union

46 Owner being male increases in magnitude yet remains insignificant when the share of females is omitted.

32

and non-union firms), yet to answer the above question of whether social security is related to union

membership, I again restrict the sample to firms which have trade unions. The results (reported in

Appendix Table G) show that for both the unbalanced and balanced panel being a union member

remains significantly positively associated with receiving social security, when all controls are

included. As for the other variables, they are as expected (not reported). Thus, within unionized

firms workers than are union members are more likely to receive social benefits than non-members.

As mentioned in section 3, in Vietnam it is more common for the union chair to be a manager or a

human resource staff member, rather than a senior worker (summary statistics are provided in

Appendix Table A). When splitting the union variable into the different union chair categories

(results not reported, but available upon request) I find that both in terms of firm and individual

wages the observed union (total) wage gap is driven solely by the case where the chair is a manager,

whereas wages in firms that have a senior worker chair (or one of the other types) are not

significantly higher than in non-union firms (results not reported). This may be related to managers

having the power to exert pressure on those who set the wages (the central authorities).

Summarizing all the results, firm unionization is associated with a higher wage bill, a result which is

particularly robust when including non-wage benefits in the wage measure, compared with a basic

measure of wages. At the individual level, however, a basic wage premium associated with individual

union membership is revealed, when comparing workers in union and non-union firms. This

difference between firm and individual level basic wages may be related partly to the fact that the

wage bill of a union firm includes the wages of non-members, which although they are not

statistically different from those of union members (as seen when the worker sample is restricted to

unionized firms), pull down the total wage. Within unionized firms the wage differential at the

individual level is not well determined, yet union membership is associated with receiving social

benefits, and this holds when restricting the sample to unionized firms only, which is in line with the

firm level results. Aside from the positive finding that being member of a trade union is associated

with receiving social insurance, the fact that a large (although declining) share of the eligible

workforce is not covered by social insurance is worrying, especially in light of the increased

regulatory effort (Law on Social Insurance) towards enhanced coverage.

33

6.3. Sensitivity analysis

(i) Location

If the existence of collective agreements is an important determinant of the union wage

association47, then given that (as noted earlier) the incidence of such agreements is considerably

higher in HCMC (65 pct. of firms) compared with in Hanoi (6 pct. of firms) (Clarke et al., (2007), a

North-South split may reveal varying results with regard to the union wage gap. Moreover, pre-

reunification Northern trade unions were characterized by closer links with the Party, whereas

Southern trade unions adopted a more adversary attitude towards the government (Edwards and

Phan, 2008), and as such were more independent. Following reunification in 1975 the system

prevalent in the North was officially adopted in the South, yet some of the ideological and

institutional differences between the two parts of the country continue to permeate through today,

and this could also be the case for trade unions.48

Further, Zhu et al. (2008) find that firms in Hanoi

remain more oriented towards (socialist) traditional personnel practices including government wage

scales, and unions’ involvement as government agents, whereas firms in HCMC have a higher rate of

adoption of modern human resource management. Given the historical and institution context

combined with the higher concentration of collective agreements in HCMC, trade unions in Southern

firms might be more accountable to their members. Table 9 shows that this seems to be the case

and that union members in Southern firms earn wages that are about 14 pct. higher than non-union

members, when both employee and firm characteristics are included.

Table 9: Individual wages

(1) (2) North South Union member 0.042 0.140*** (0.037) (0.039) Employee characteristics Yes Yes Firm characteristics Yes Yes Observations 619 534 R-squared 0.381 0.505 Note: Dependent variable: Log real individual wages. OLS. Year dummy included. Robust standard errors (in parenthesis). *** p<0.01, ** p<0.05, * p<0.1

47 The question on collective agreements was only introduced in the 2009 survey, and therefore I cannot use this information in the analysis. 48 Kim (2008) provides a captivating discussion of institutional differences between North and South Vietnam.

34

(ii) Compliance

If unions in Southern firms are more responsive to their members this could also be an indication

that Southern firms are more likely to comply with regulations in terms of providing social benefits.

Table 10 presents the Probit results where the provision of social insurance is the outcome variable

taking the value 1 if the firm provides social insurance and 0 otherwise (see Table 1 for summary

statistics). Indeed firms in Southern provinces are more likely to be comply with social insurance

regulation, and in line with the results in Table 8, firms with trade unions are more likely to provide

social benefits. Moreover, in addition to firm size compliers and non-compliers differ along the legal

status, professional and female share dimensions (not reported). Finally, the year dummy indicates

that compliance has improved over time, as also shown in Table 1.

Table 10

Probit Trade union 0.605*** (0.056) Firm size (log employment) 0.170*** (0.054) South 0.329*** (0.078) Owner male Yes Owner education

Yes

Legal dummies Yes Worker shares Yes Province dummies Yes Sector dummies Yes Year dummy 0.116* (0.068) Observations 337 Note: Dependent variable: firm provides social insurance. Probit estimates, marginal effects. Robust standard errors (in parenthesis) .*** p<0.01, ** p<0.05, * p<0.1

7. Conclusion

This paper has examined the union wage differential among Vietnamese small and medium

manufacturing enterprises using matched employer-employee survey data from 2007 and 2009.

Vietnam provides an interesting case study due to the transitional environment in which unions

operate, and the fact that union membership, by contrast to many developed countries, is on the

35

rise. The detailed nature of the dataset has allowed for addressing some of the methodological

concerns with this kind of analysis. Moreover, information on individual union membership has

permitted the separation of the individual union wage gap from the firm union wage differential.

Finally, the fact that not all firms are unionized has provided an opportunity to more carefully

analyze the individual wage gap.

The results show that unionized firms have higher wage bills, when controlling for standard wage

determinants, and the result is particularly robust when other labour costs are included in the wage

measure. At the individual level, the results reveal a basic wage premium for union members, yet

not when comparing union and non-union workers within unionized firms only. In the latter case,

however, union membership is strongly associated with receiving social benefits, in accordance with

the firm level results. All in all, the results suggest that Vietnamese trade unions play an important

role in ensuring the provision of social benefits. Although this is a positive finding in terms of the

effectiveness of unions in protecting the rights of their members, it should not be the responsibility

of unions to ensure that firms comply with regulations. Unionized or not, firms that are formally

registered are mandated by the Law on Social Insurance (2007) to contribute towards social security

for workers (employed on contracts of 3 months or more). Thus, one policy implication arising from

this study is the need to improve the enforcement of regulations, while at the same time enhancing

workers’ awareness of their rights. Given that Southern firms appear to be more likely to comply

with regulations, they could potentially act as role models in this regard.

Another finding that emerges from the analysis is that lower ability workers are less likely to be

union members, because they sort into non-union firms, which are characterized by lower

compensation and benefits. In the presence of a “threat effect” non-union employers would feel

threatened by their workers potentially becoming unionized, in which case total wages between

union and non-union firms should be more comparable. However, given that a large share of firms

remain non-unionized, and the pressure on them to set up unions remains limited, unions do not

seem to pose a real “threat”. Thus, also in this regard there is a need for enhanced regulatory

enforcement so as to ensure that unions are established, and strive to live up to their mandate of

representing the interests of all workers.

36

8. References

Abowd, J. M., and Kramarz F. (1999). Chapter 40 “The analysis of labor markets using matched employer-employee data”. Handbook of Labor Economics, Elsevier. Volume 3, Part 2: 2629-2710.

Acemoglu D., Aghion P., Violante G., (2001), “Deunionization, technical change and inequality”, Carnegie-Rochester Conference Series on Public Policy, Vol. 55 (1), pp. 229-264

Benson, J. and Gospel, H. (2008). “The emergent enterprise union? A conceptual and comparative analysis”, The International Journal of Human Resource Management, Vol. 19 (7), pp. 1365–1382.

Booth, A.L. and Chatterji, M. (1995) “Union Membership and Wage Bargaining when Membership is not Compulsory”, The Economic Journal, Vol. 105 (429), pp. 345-360.

Booth, A.L. (1995), The Economics of the Trade Union, Cambridge University Press, Cambridge.

Boeri, T. van Ours, J. (2008), The Economics of Imperfect Labor Markets, Princeton University Press, Princeton.

Brugiavini et al. (2001). What do Unions Do to the Welfare States? A Report for the Fondazione RODOLFO DEBENEDETTI.

Card, D. & De la Rica, S. (2006), “Firm-level contracting and the Structure of Wages in Spain”, Industrial and Labor Relations Review, Vol. 59, No. 4, pp. 573-592.

CIEM , DOE, ILSSA (2010) Characteristics of the Vietnamese Business Environment: Evidence from a SME survey in 2009. CIEM, Hanoi.

Clarke, S., Lee, C-H. and Chi, D.Q. (2007) “From rights to interests: The Challenge of Industrial Relations in Vietnam”, Journal of Industrial Relations, Vol. 49 (4), pp. 545-568.

Croson, R. and Gneezy, U. (2009). ‘Gender Differences in Preferences’, Journal of Economic Literature, 47 (2), pp. 1-27.

Dell’Aringa, C. & Lucifora C. (1994), “Collective Bargaining and Relative Earnings in Italy”, European Journal of Political Economy, Vol.10, pp.1001-44.

Edwards, V. & Phan, A. (2008), “Trade Unions in Vietnam: From socialism to market socialism” in Benson, J. & Zhu, Y. Trade Unions in Asia: An economic and sociological analysis, Routledge: New York.

GSO (2008). The real situation of enterprises through the results of surveys conducted in 2005, 2006, 2007, Statistical Publishing House: Hanoi.

Freeman, R.B. & Medoff, J.L. (1984) What do Unions do? Basic Books, Inc. Publishers, New York.

GSO (2004). Results of establishment census of Vietnam 2002, Volume 2 -Business Establishments, Statistical Publishing House: Hanoi.

Hildreth, A. K. and Pudney S. E. (1997), "Employers, workers and unions: an analysis of a firm worker panel with endogenous sampling, attrition and missing data" ,Working paper (University of Essex).

ILO (2010), Database of International Labour Standards http://www.ilo.org/ilolex/english/convdisp1.htm

ILO (2008).Global Wage Report 2008/09. International Labour Office: Geneva. ILSSA/ MoLISA (2010). Note on Trade Union Law (1990). ILSSA/ MoLISA: Hanoi. Jones, P. (2001). ‘Are educated workers really more productive?’ Journal of Development Economics,

vol. 64, pp.57-79. Kim, A.M. (2008). “Learning to Be Capitalists: Entrepreneurs in Vietnam’s Transition Economy”, Oxford

University Press. Krueger, A. and Summers, L.H.(1988), “Efficiency wages and the Inter-Industry Wage Structure”,

Econometrica, Vol.56, No-2 (March), pp.259-93.

37

Larsen, A.F., Rand, J. and Torm, N. (2009). “Do Recruitment Ties Affect Wages? An Analysis Using Matched Employer-Employee Data from Vietnam”, Review of Development Economics, forthcoming.

Lalonde, R. Marschke , J. G. and Troske, K. (1996), "Using longitudinal data on establishments to analyze the effects of union organizing campaigns in the United States", Annales d'economie et de statistique 41-42: 155-186.

Lewis, H.G. (1986) “Union relative wage effects” Handbook of Labor Economics Volume 2

Liu, A. Y.C.

, 1986, Pages 1139-118.

(2004)

Lucas, R. (1988). ‘On the Mechanics of Economic Development’, Journal of Monetary Economics, 22, pp.3-39.

. ‘Changes in gender wage gap in Vietnam’, Journal of Comparative Economics, Vol. 32(3) pp. 586-596.

Malesky, E. (2009). The Vietnam Provincial Competitiveness Index: Measuring Economic Governance for Private Sector Development. Vietnam Chamber of Commerce and Industry (VCCI and United States Agency for International Development’s Vietnam Competitiveness Inititative (VNCI): Ha Noi, Vietnam.

Mincer, J. (1974). Schooling, Experience and Earnings, NBER and Columbia University Press. Nguyen, B., Loi, C.C and Nguyen, C.H. (2006) “A stocktaking of Viet Nam’s Labor Market Policies”, in

Felipe, J. & Hasan R. (eds.) (2006) Labor Markets in Asia: Issues and Perspectives.

Nickell, S. and Layard, R. (1999). “Labor market institutions and economic performance”, Handbook of Labor Economics, Volume 3, Part 3.

London: Palgrave Macmillan for the Asian Development Bank.

Oi, W. and Idson, T.(1999). ‘Firm size and wages’, in Aschenfelter O. and Card, D. (eds.), Handbook of Labor Economics, 3, pp.2165-2214.

Park, Y-B. (1991), Union/Nonunion Wage Differentials In the Korean Manufacturing Sector, International Economic Journal, vol. 5 (4), pp. 79-91.

Pencavel, J. (2005). “Unionism Viewed Internationally”, Journal of Labor Research, Volume XXVI (1). Phan, D. (2009). Background calculations to “A report on Vietnam’s Labor Market”. Prepared for the

CIEM-Danida Project. CIEM: Hanoi. Quang, V. D. (2008). “Viet Nam: workers in transition” in Lee, S. and Eyraud, F. (eds.), Globalization,

flexibilization and working conditions in Asia and the Pacific, ILO: Geneva. Rama, M. (2000), Wage misalignment in CFA countries: were labour market policies to blame? Journal

of African Economies, vol. 9 (4), pp. 475-511. Rand, J. and Tarp F. (2011). "Does Gender Influence Fringe Benefits Provision? Evidence from

Vietnamese SMEs”, Rosenbaum, J., DeLuca, Miller S.R. and Roy, K. (1999). ‘Pathways into work: ‘Short and Long-terms

effects of personal and Institutional ties’, Sociology of Education, vol. 72, pp. 179-196.

Feminist Economics.

Schultz and Mwabu (1998). Labour Unions and the Distribution of Wages and Employment in South Africa. Industrial and Labor Relations Review. 51, pp. 680-703.

Söderbom, M., Teal. F., Wambugu, A. (2005). ‘Unobserved heterogeneity and the relation between earnings and Firm Size: evidence from two developing countries’, Economics Letters 87(2), pp. 153-59.

Spence M. (1973). ‘Job Market Signaling’. Quarterly Journal of Economics 87 (3): 355–374.

Tybout, J. R. (2000), “Manufacturing Firms in Developing Countries: How Well Do They Do, and Why?” Journal of Economic Literature

Yoon, Y. (2009), A comparative study on industrial relations and collective bargaining in East Asian countries, Working paper, ILO: Geneva.

, Vol. 38 (1), pp. 11-44.

VGCL (2010), Steering Committee of Trade Union Members Development, VGCL: Hanoi. 20 May 2010. Zhu, Y., Collins, N., Webber, M. and Benson, J. (2008), “New forms of ownership and human resource

practices in Vietnam”, Human Resource Management, vol. 47 (1), pp.157-175.

38

Zhu, Y. and Fahey S. (2000). 'The Challenges and Opportunities for the Trade Union Movement in the Transition Era: Two Socialist Market Economies - China and Vietnam'. Asia Pacific Business Review, Special issue on ‘Globalization and Labour Market Deregulation: Trade unions in the Asia Pacific Region. Vol. 6, Nos. 3&4, pp282-299.

APPENDIX: For both firm and employee wages, outlier tests were done with the dual purpose of removing measurement

errors and large outliers, which might otherwise be driving the results. Given the enormous variation in wages

between different job categories employee wages were cleaned based on the average figures for different

occupations, as provided in Quang (2008). Observation with wages above (or below) 3 standard deviations

were removed from the sample and other suspicious observations were scrutinized. Furthermore the reported

individual wage was compared with the average wage in the firm, revealing consistency on both sides. Both

the firm and employee wages were deflated using the national CPI for the relevant years. Figures 1a and 1b

show the kernel density for firm (total) wages and individual wages, based on the unbalanced panels of firms

and workers. The balanced panel densities (not reported) are similar and available upon request.

Figure 1a: Density of firm wages (including other labour costs)

0.1

.2.3

.4Ke

rnel

den

sity

8 9 10 11 12 13Log real monthly wage

Unionized firms Non-unionized firms

39

Figure 1B: Density of employee wages

0.2

.4.6

.81

Ker

nel d

ensi

ty

5 6 7 8Log real monthly wage

Union members Non-union members

Table A: Summary Statistics of Additional Firm Variables

Firm sample (Unbalanced)

Employee sample (Unbalanced)

Wage share of total value added (logged) 0.368 0.366

Union chair (as a share of unionized firms):

(i) Senior worker 0.138 0.156

(ii) Manager 0.394 0.415

(iii) Human resource person 0.339 0.306

(iv) Owner of the enterprise 0.073 0.054

Ha Noi (Yes = 1, No = 0) 0.204 0.219

Phu Tho (Yes = 1, No = 0) 0.030 0.023

Ha Tay (Yes = 1, No = 0) 0.083 0.080

Hai Phong (Yes = 1, No = 0) 0.124 0.132

Nghe An (Yes = 1, No = 0) 0.080 0.083

Quang Nam (Yes = 1, No = 0) 0.065 0.045

Khanh Hoa (Yes = 1, No = 0) 0.086 0.074

Lam Dong (Yes = 1, No = 0) 0.035 0.042

HCMC (Yes = 1, No = 0) 0.246 0.263

Long An (Yes = 1, No = 0) 0.047 0.040

South (Yes = 1, No = 0) 0.479 0.463

Sector 1 Food products and beverages 0.169 0.197

Sector 2 Tobacco products 0.000 0.000

Sector 3 Textiles 0.062 0.062

Sector 4 Wearing apparel etc. 0.065 0.062

Sector 5 Tanning and dressing leather 0.009 0.004

Sector 6 Wood and wood products 0.098 0.087

Sector 7 Paper and paper products 0.068 0.066

Sector 8 Publishing. printing etc. 0.059 0.066

Sector 9 Refined petroleum etc. 0.003 0.005

40

Sector 10 Chemical products etc. 0.041 0.048

Sector 11 Rubber and plastic products 0.104 0.105

Sector 12 Non-metallic mineral products 0.027 0.027

Sector 13 Basic metals 0.024 0.023

Sector 14 Fabricated metal products 0.121 0.109

Sector 15 Electrical machinery and office machinery etc. 0.059 0.052

Sector 16 Vehicles etc. 0.009 0.016

Sector 17 Transport equipment 0.012 0.015

Sector 18 Furniture, jewellery and musical instruments etc. 0.071 0.056

Sector 19 Recycling 0.000 0.000

Total number of observations 338 1153

Table B: Summary statistics of additional (Firm) variables (unbalanced panel)

Total 2007 2009 Mean SD Mean SD Mean SD Firm size (regular full-time employees) 40.386 39.640 38.734 35.686 41.552 42.195 Private firm = 1 0.209 0.409 0.218 0.413 0.203 0.402 Cooperative/Collective/Partnership = 1 0.090 0.287 0.107 0.309 0.078 0.269 Limited liability = 1 0.600 0.490 0.604 0.490 0.598 0.491 Joint stock company = 1 0.101 0.301 0.071 0.258 0.121 0.327 Owner is male 0.610 0.488 0.614 0.487 0.607 0.489 Owner has higher education 0.909 0.288 0.899 0.301 0.916 0.280 Professional share 0.093 0.075 0.097 0.078 0.091 0.072 Female share 0.416 0.235 0.414 0.234 0.418 0.235 Temporary worker share 0.114 0.181 0.120 0.198 0.109 0.167 Location (Urban = 1, Rural = 0) 0.614 0.487 0.635 0.482 0.600 0.490 North (North = 1, South = 0) 0.537 0.499 0.570 0.496 0.513 0.500

Total number of observations 1153 477 676

Table C: Union membership determinants Probit Male -0.017 (0.033) Age 0.030*** (0.011) Age squared -0.030** (0.014) Higher education 0.041 (0.043) Manager -0.007 (0.050) Professional 0.042 (0.051) Sales 0.067 (0.060) Other 0.007 (0.044) Informal 0.002 (0.033) Firm size (regular full-time employees) 0.336*** (0.022)

41

South 0.085** (0.033) CCP 0.511*** (0.065) Limited liability 0.081* (0.049) Joint stock company 0.034 (0.072) Owner is male 0.021 (0.032) Owner has higher education 0.174*** (0.049) Professional share -0.095 (0.226) Female share -0.000 (0.085) Temporary worker share -0.002 (0.098) Observations 1148 Note: Dependent variable: worker is union member. Probit estimates, marginal effects. Year dummy and sector dummies included. Robust standard errors (reported in parenthesis). *** p<0.01, ** p<0.05, * p<0.1.

Table D: Firm basic wage Unbalanced Panel Balanced Panel VARIABLES (1) (2) (3) (4) Trade union 0.125 0.132* 0.103 0.208** (0.081) (0.073) (0.099) (0.100) Firm size 1.015*** 1.033*** 0.953*** 0.946*** (0.039) (0.040) (0.051) (0.058) Owner male No Yes No Yes Owner education No Yes No Yes Legal dummies No Yes No Yes Worker shares No Yes No Yes Province dummies No Yes No Yes Sector dummies No Yes No Yes R-squared 0.770 0.842 0.738 0.838 Observations 338 338 188 188 Note: Dependent variable: Log real basic wages. OLS. Year dummy included. Robust standard errors (in parenthesis). *** p<0.01, ** p<0.05, * p<0.1

Table E: Wage share of real value added

Unbalanced Panel Balanced Panel (1) (2) OLS OLS Union firm 0.006 0.012 (0.019) (0.026) Firm size 0.020** 0.015 (0.010) (0.017) Firm characteristics Yes Yes Province dummies Yes Yes Sector dummies Yes Yes R-squared 0.231 0.321 Observations 335 185 Note: Dependent variable: Log wage share of real value added. OLS. Year dummy included. 3 observations are dropped due to share > 1. Robust standard errors (in parenthesis). *** p<0.01, ** p<0.05, * p<0.1

Table F: Individual wages in unionized firms

Unbalanced Panel Balanced Panel (1) (2) OLS OLS

Union member 0.074 0.184* (0.049) (0.104)

42

Employee characteristics Yes Yes Firm characteristics Yes Yes

R-squared 0.535 0.741 Observations 476 155

Note: Dependent variable: Log real individual wages. OLS. Year dummy included. Robust standard errors (in parenthesis) clustered by worker id (for the balanced sample). *** p<0.01, ** p<0.05, * p<0.1

Table G: Social insurance in unionized firms

Unbalanced Panel Balanced Panel (1) (2) (3) (4) Probit Probit Probit Probit Union member 0.267*** 0.817*** 0.296** 0.771*** (0.063) (0.094) (0.142) (0.128)

Employee characteristics Yes Yes Yes Yes Firm characteristics No Yes No Yes

Observations 476 377 155 93 Note: Dependent variable: worker receives social insurance. Probit estimates, marginal effects. Year dummy included. Robust standard errors (reported in parenthesis). *** p<0.01, ** p<0.05, * p<0.1. The large drop in the number of observations between columns 1 and 2, and 3 and 4 is due to multicollinearity among certain sectors.