Upload
kathlyn-jennings
View
214
Download
0
Embed Size (px)
Citation preview
Disability questions in censuses and surveys
Methodological issues: proxy response, non-response and
mode of administration
21-23 September 2005
Montserrat López-Cobo
Fifth Meeting of the Washington Group
Fifth Meeting of the WASHINGTON GROUP 2005
Are proxy-responses different from self-responses? Why?
Do all proxies provide the same quality responses? Who is the best proxy?
Is there an effect on the quality of data? Can we measure it? Can we correct it?
Proxy response
Fifth Meeting of the WASHINGTON GROUP 2005
It is generally accepted that proxy- and self- responses differ
Many studies found that proxy-respondents underreport disability-related information: Disability Impairments of emotion and Pain Limitation of activities, long-term
disability, need for assistance with ADL Chronic condition, days of activity
restriction, bed disability days General health events and conditions
Proxy responseLevel differences 1
Fifth Meeting of the WASHINGTON GROUP 2005
... While some other studies found that proxies overreport: Cognitive status, Mild Cognitive
Impairment ADL disability Functional health, when the self-
respondent is demented
Proxy responseLevel differences 2
Fifth Meeting of the WASHINGTON GROUP 2005
Differences in cognitive procedures (1/2) Type and level of available information
o Proxies report impairments or conditions that are more observable or mentioned to them
o Proxy reports are more accurate for conditions that are serious, painful, persistent or potentially life threatening
If person under 65: proxies underreportIf person 65 and over: proxies overreport
Proxy information is more stable. Self information is more dynamic Self-responses are less consistent across time
Proxy responseUnderstanding self-proxy differences 1
Fifth Meeting of the WASHINGTON GROUP 2005
Differences in cognitive procedures (2/2) Proxies engage in less extensive cognitive
process. Provide heuristic-based responses Proxies rely more on inferences and estimations.
Proxies overreport disabilities aparently related to a previously reported disability and underreport disabilities seemingly unrelated to the previously reported.
Proxies are less affected by social desirability concerns Proxies are more willing to report disability
Proxy responseUnderstanding self-proxy differences 2
Fifth Meeting of the WASHINGTON GROUP 2005
Social relationship with the selected person The closest the relationship, the most
accurate the proxy-response Spouse’s responses are better than other
proxies’ Proffesionals and caregivers provide more
accurate information than lay proxies
Length of relationship Positive effect in convergence self-proxy
Memory capacity of respondent Affect to data quality
Proxy responseThe best proxy-respondent
Fifth Meeting of the WASHINGTON GROUP 2005
Types of methods: 1) Traditional
First step: Measure the bias by comparing proxy- estimates with self- estimates
Second step: Introduce statistical control adjusting for demographic or health-related variables
2) New strategy First: Set assumptions about the nature of
differences between self- and proxy-responses
Second: Model and estimate these differences based on the assumptions
Proxy responseMeasuring and correcting bias
Fifth Meeting of the WASHINGTON GROUP 2005
Mode of administration
Fifth Meeting of the WASHINGTON GROUP 2005
Differences in the sample Population coverage Response rates
Differences due to social context within which data are collected Contact with interviewer
Non-verbal cuesSocial desirability bias: Reduced with
self-administered questionnaires
Mode of administrationSources of differences among modes 1
Fifth Meeting of the WASHINGTON GROUP 2005
Differences intrinsic to the mode Visual materials cannot be used by phone Complexity of the questionnaire is limited
by the mode Stimuli in a visual mode Primacy effects
Stimuli in a hearing mode Recency effectsDifferences in responses involving long
lists Order effects Effects of the mode on the interviewer
Mode of administrationSources of differences among modes 2
Fifth Meeting of the WASHINGTON GROUP 2005
Telephone interviews Do not underrepresent people with disabilities Sometimes is not as well accepted by population
as face-to-face interviews Proved useful to assess mental health using
recommended instruments (GHQ-12, CIS-R, CIDIS), the Expanded Disability Status Sacle
Telephone vs Mail Mail responses report poorer health and more
chronic conditions than Telephone Differential non-response rates by age:
Elder: NR (Telephone) > NR (Mail) Young: NR (Telephone) < NR (Mail) General: Item NR (Telephone) < Item NR
(Mail)
Mode of administrationResults from comparative studies 1
Fifth Meeting of the WASHINGTON GROUP 2005
Face-to-Face vs Mail Health differences between
respondents and non-respondents by mode: non-response bias Non-respondents and late
respondents to Mail are more cognitively impaired and more disabled than respondents (among elderly).
Non-respondents to F-t-F are similar to respondents.
Mode of administrationResults from comparative studies 2
Fifth Meeting of the WASHINGTON GROUP 2005
Telephone/CATI vs Face to Face Differential non-response rates:
NR (CATI) > NR (F-t-F) Measurement bias:
MBias (CATI) < MBias (F-t-F) No differences between CATI and F-t-F
for reports on chronic conditions, activity limitations and disability rates
CATI vs CAPI Differential non-response rates by age:
Elder: NR (CATI) > NR (CAPI)Young: NR (CATI) < NR (CAPI)
Mode of administrationResults from comparative studies 3
Fifth Meeting of the WASHINGTON GROUP 2005
CASI Reduces social desirability bias Enhances the feeling of privacy Respondents generally like CASI Technological possibilities have a
positive influence on data quality (minimizing errors)
Respondent’s self-disclosure is higher in CASI
Mode of administrationResults from comparative studies 4
Fifth Meeting of the WASHINGTON GROUP 2005
Disabilities: Physical, sensory, mental illness and mental retardation
Challenges addressed: communication, fatigue and cognitive issues
Questionnaire design Eliminate soft consonant sounds (s, z, t, f and g)
to overcome high-frequency hearing loss Build in "breaks" for respondents to let them rest Incorporate neutral encouragement to avoid
drop-outs Design checks for unexpected responses Use structured probes for questions that might
be difficult to understand
Mode of administrationAn experience interviewing disabled people by telephone
Fifth Meeting of the WASHINGTON GROUP 2005
Interviewer training and supervision Usual background and purpose of study Training on challenges likely to face Sensitive exercise regarding the treatment
to the disabled person Guidance to overcome each of the
challenges Support to interviewers and reduction of
stress
Other recommendations Interviews take longer Multiple sessions
may be required
Mode of administrationAn experience interviewing disabled people by telephone
Fifth Meeting of the WASHINGTON GROUP 2005
Non-response
Fifth Meeting of the WASHINGTON GROUP 2005
Components of non-response Non-contact Non-cooperation
Non-contact: the household pattern of disabled Are older and likely to live in non-
metropolitan areas higher probability of being contacted
Less likely to live with children lower prob. of contact
May be more fearful of opening doors to strangers lower prob. of contact
Non-response and Disability
Fifth Meeting of the WASHINGTON GROUP 2005
Non-cooperation: exchange theory vs social isolation theory Exchange theory: Persons who feel that the
survey sponsor has provided (or could provide) benefits to them are more likely to cooperate
Persons with disabilities might be expected to be more cooperative with a government-sponsored survey
Social isolation theory: People who are isolated from the mainstream society feel less responsibility toward government and are less likely to cooperate
Persons with disabilities might be expected to be less cooperative
Non-response and Disability
Fifth Meeting of the WASHINGTON GROUP 2005
Persons with severe disability are more likely than persons with less severe disability to be contacted and to cooperate, but when they are interviewed they are less likely to answer for themselves; instead, proxy and assistant respondents tend to answer for them.
Elder people with disabilities respond sooner than young and non-disabled to mail surveys. These findings support Exchange theory
Self-perceived memory problems increase item non-response and “I don’t know” answers.Not preceived memory deficits implies inaccurate information on items requiring recall.
Non-response and DisabilityResults from studies 1
Fifth Meeting of the WASHINGTON GROUP 2005
Health of non-respondents is worse than respondents’ in terms of: stroke, Basic ADL, mobility disabilities, self-rated health and mortality rates.
Respondents with poorer physical functioning and/or limiting long-term illness have higher non-response rates. These findings support Social isolation
theory
Non-response and DisabilityResults from studies 2
Fifth Meeting of the WASHINGTON GROUP 2005
Recommendations for including disabled people in
interview surveys
Fifth Meeting of the WASHINGTON GROUP 2005
Bias due to exclusion is amplified in surveys where disability is a key measure of interest
Recommendations: Include institutionalised population
in samples Carry out the interview in a private
environment Provide adaptative technologies and
procedures (of primary importance in visual, hearing and speech impairment).
Use simple and clear questions
Fifth Meeting of the WASHINGTON GROUP 2005
Recommendations: Special training for interviewers Provide alternative modes of
administration If proxy is used:
Respondent’s assessments of proxy’s answer is valuable
The proxy should be nominated by the selected respondent
Fifth Meeting of the WASHINGTON GROUP 2005
Proxy response validity depends on factors such as: proxy choice and distinction, topic investigated, health condition of the selected person...
Mode of administration (unique or a mix-mode) should be decided taking into account: topic investigated, population objective of the survey, expected acceptability of the mode by the population, technological possibilities...
Final conclusion
Fifth Meeting of the WASHINGTON GROUP 2005
Different patterns of response might be indicative of bias which can affect estimates.
Non-response is one of the sources of data error. But not the unique... Emphasis should be given not only to minimazing non-response rates but also to estimation and control of measurement error.
Final conclusion
21-23 September 2005
Fifth Meeting of the Washington Group