Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Hello? is better than Hello: Effects ofgreeting intonation on participation in
survey invitationsJosé Benkí,1 Jessica Broome,1 Frederick Conrad,1,2
Robert Groves,3 and Frauke Kreuter2
1 University of Michigan2 University of Maryland3 Georgetown University
AAPOR 68th Annual Conference, BostonMay 2013
Acknowledgements• Support:
– NSF Grants # SES-0819734 and # SES-0819725 ;– Survey Research Center, University of Michigan– Dept. of Communicative Sciences & Disorders, Michigan
State University
• Coding, Transcribing, Acoustic Analysis, Sampling– Pete Batra, Rachel Benner, Kelly Franckowiak, Haley Gu,
Ben Jarvi, Emily Kordupel, Peter Kotvis, Abby Lincoln, LacieLinstrom, Melissa Littlefield, Daniela Lopez, ColleenMcClain, Joe Matuzak, Colleen McCarty, Patty Maher, GabeMoss, Kirsten Mull, Danny Nielsen, Dana Perkins, FernandoPacheco, Danielle Popielarz, Christine Sheffler, AmandaTatro, Dave Vanette, and Dylan Vollans
Voice, language, participation, andgreetings
• Telephone interviewers’ success obtaining interviews is due, atleast in part, to what they communicate over the phone
– How they Sound -- vocal attributes– How and What they Speak -- manner and content of speech– How they Interact with potential respondents
• Over their careers, some Iwers are more and others lesssuccessful; this strongly implies that differences in their verbalattributes play an important part in outcomes
• At the same time, potential respondents quickly make aparticipation decision, often within 30 seconds or less.
• The initial portion of the survey invitation, the greeting, is likely toplay a critical role in the participation decision.
Example “Hello” pairsInvitation 1:
Invitation 2:
Invitation 3:
Invitation 4:
Note that we are focusing in thispaper on the greeting and decision toparticipate, not on the interview itself.
Attributes of a successfulgreeting: survey methodology
• Survey methodology literature has considered invitationsfrom a speech perspective but not greetings specifically(Oksenberg, Coleman & Cannell, 1986; Oksenberg &Cannell, 1988)
• Higher pitch (Sharf & Lehman, 1984; Groves, O’Hare,Gould-Smith, Benki, Maher 2008)
• Lower pitch for male interviewers (Benkí, Broome,Conrad, Groves, and Kreuter 2011)
• Less scripted, more extemporaneous deliveries (Groves,et al. 2008), perhaps mediated by moderate levels ofdisfluency, rate, and pausing (Conrad, et al. 2013; Benkí,Broome, Conrad, Groves, and Kreuter 2011)
Attributes of a successful greeting:conversation and pragmatics
• “Prosodically-large” face-to-face greetings (Pillet-Shore 2012)
Signal familiarity, approval, andthe special status of themeeting
LongerOverlappingLouderHigher pitchWider pitch range
• High-pitched greetings intelephone conversations signalenthusiasm, recognition, andfriendliness (Schegloff 1998)
• “Prosodically-small” face-to-face greetings
Signal unfamiliarity, a neutralstance (at best), that theaddressee has already beengreeted, and potential fordisapproval.
ShorterConsecutiveSofterLower pitchNarrower pitch range
6
Focusing on “Hello” in thesurvey invitation
• The “hello” is one of the first sources ofinformation the householder receives inmaking her participation decision
• The “hello” greeting is the most frequenttelephone greeting in the US
• Restricting the study to this single utterancewithin the invitation allows us to control forlinguistic content while reducing the samplesize only moderately.
7
Our Project• Examines impact of Iwers’ voice, speech and interaction
with phone answerer on decision to– Participate (Agree)– Refuse to Participate (Refuse)– Defer (Scheduled Callback)
• Measures:– Transcription and turn marking– Acoustic (e.g., pitch or f0, speech rate, vowel quality)– Paralinguistic (e.g., fillers, i.e., uh, um)– Move coding (e.g., self-identification, duration question )– Global ratings: (e.g., animation, similarity of I and A accent)
• 1380 audio recorded telephone introductions from 5studies at University of Michigan SRC
• 100 interviewers
Multilevel Data Set
Contact
Turn
Move
Move Move
Turn Turn
Interviewers
Study
- Interviewer is cross-classified with study.
-Interviewer is cross-classified withcases/samples.
-Each level is completelynested in the subsequentupper level.
Cases
1380 contacts, transcribed andcoded
Transcription and acousticanalysis
Hypothesis: High pitchedgreetings lead to participation
Householder• Friendliness• Recognition• Approval
Interviewer• Enthusiasm• The householder
and/or her concernsare special to theinterviewer (Groves,Singer, & Corning2000)
11
Distinguishing between higher overallpitch and higher pitch in the greeting
• Previous work by our group (AAPOR2011) has documented an inverseassociation between participation andthe overall pitch of male interviewers(n.s. trend for female interviewers)
• Here we focus specifically on the pitchduring “Hello”.
12
Normalizing by the speaker’sbaseline pitch
For a talker Ti in contact i and turn j where T can beeither householder (H) or interviewer (Iwer), Praatmeasures the following values:
F0o(Ti) = median(F0) during the /o/ of the “hello” of Ti
F0j(Ti) = median(F0) during turn j of Ti
From the turn- and contact-level datasets, we computethe following contact-level values, where N is the totalnumber of turns j in contact i for each Ti:
F0baseline(Ti) = Σj F0j(Ti) ⁄ NF0delta(Ti) = [ F0o(Ti) – F0baseline(Ti) ] ⁄ F0baseline(Ti)
13
Hypotheses
• F0delta is a signed measure of pitch change in the(hell)“o” of the greeting, normalized by F0baseline .
• F0delta(Householder) will be positively correlated withparticipation.High-pitched HH greetings signal approval and
potential willingness to participate• F0delta (Interviewer) will be positively correlated
with participation.High-pitched Iwer greetings signal enthusiasm
and that the HH is special to the interviewer
14
Sample for analysisOutcome All contacts Iwer hello HH hello Both hellosRefusal 566 (41) 290 (40) 206 (33) 199 (32)Agree 263 (19) 125 (17) 125 (20) 124 (20)SCB 537 (39) 301 (42) 294 (46) 291 (47)Other 14 (1) 8 (1) 8 (1) 8 (1)TOTALCONTACTS
1380 724 633 622
15
•KEY: Counts (Column percentages)•All contacts (column 1): 100 interviewers, 671households•Both hellos (column 4): 86 interviewers, 426households
F0delta (Interviewer ) for agrees (F[1, 87] = 4.21; Prob > F = 0.0433)and SCBs (F[1, 87] = 6.82; Prob > F = 0.0106) are significantly higher thanrefusals (S.E.s are adjusted for clustering by interviewers)F0delta (Householder) differences by outcome are n.s.
16
Evaluating for interaction betweenhouseholder and interviewer hello
• Interviewers do not appear to match the intonationpattern of the householder greeting:r [F0delta(Iwer), F0delta(HH)]=0.0651, p>0.10
• Nevertheless, the householder greetingprecedes the interviewer greeting
• We can analyze the householder greetingintonation as a context or condition on theinterviewer greeting
17
Evaluating for interaction betweenhouseholder and interviewer hello
• Contacts were classified according towhether F0delta(Ti) exceeded the corpusmeans.
• Means (SD): F0delta (Iwers)=0.236 (0.306)F0delta (HH)=0.215 (0.332)
• Four categories according to HH andInterviewer F0delta (intonation on the hello):HH&I High I>HH Iwer high, HH lowHH&I Low HH>I HH high, Iwer low
18
Interaction of hellosHello intonation pattern TOTAL
HH&I Low I>HH HH>I HH&I HighRefusal 64 (34)
0.145 (29)0.5
53 (42)3.8
37 (26)1.9
199 (32)6.4
Agree 40 (21)0.1
35 (23)0.4
15 (12)4.2
34 (24)0.9
124 (20)5.6
SCB 87 (46)0.1
75 (48)0.0
57 (46)0.1
72 (50)0.3
291 (47)0.5
TOTAL 191 (100)0.3
155 (100)1.0
125 (100)8.1
143 (100)3.1
614 (100)12.4
19
Key: Frequency (column percentage) chi2 contributionPearson chi2(6)=12.4228 Pr =0.053
Logistic regression models ofnonrefusal
• Cross-classified multilevel logistic with randomeffects of interviewer and household (xtmelogit inStata 12)
• Model 1: F0delta(Iwer) All contacts• Model 2: F0delta(Iwer) Contacts with low-
pitched householder greeting (columns 1 & 2)• Model 3: F0delta(Iwer) Contacts with high-
pitched householder greeting (columns 3 & 4)• Model 4: F0delta(Iwer), Householder greeting
pattern (HHigh), + interaction, All contacts 20
Logistic regression models ofnonrefusal
Predictor Model 1Coeff (SE)
Model 2 Model 3 Model 4
Constant 0.763** (0.162) 0.817** (0.183) 0.442* (0.211) 0.922** (0.201)
F0delta(Iwer) 0.763* (0.366) 0.306 (0.423) 1.398* (0.547) 0.277 (0.474)
HHigh -0.398 (0.271)
F0delta(Iwer) x HHigh 1.165 (0.734)
N 622 351: HHigh=0 271: HHigh=1 622
Goodnessof fit
χ2=4.34, df=1,p=0.037
χ2=0.53, df=1,p=0.468
χ2=6.53,df=1,p=0.011
χ2=6.95,df=3,p=0.073
Random effects
Interviewer 0.421 (0.152) 0.394 (0.206) 0.419 (0.268) 0.424 (0.153)Household 0.924 (0.249) 0.406 (0.488) 0.656 (0.434) 0.926 (0.251)
21* p<0.05 **p<0.01
22
Discussion
• Pitch in householder hellos does notappear to be strongly correlated withparticipation
• However, householder hello intonationdoes appear to predict when a high-pitched interviewer greeting is mostimportant
23
Discussion
• Householders with low-pitched hellos showlittle preference for interviewer high-pitchedhello intonation
• Householders with high-pitched hellos doprefer high-pitched interviewer hellos
• For these householders: Participationincreases with interviewer hello pitch:30% increase in F0delta(Iwer) (~1 SD) results in a 9
point increase in nonrefusals
24
Why the interaction between householderand interviewer greeting?
• Perception follows production: thehouseholders who produce high-pitched(prosodically-large) greetings also respondto (and expect) this greeting style frominterviewers.
• For householders who produce low-pitched(prosodically-small) greetings: a low-pitchedinterviewer greeting is truly neutral (andnot negative)
25
Practical Implications
• Based on these data, interviewers should use a high-pitched greeting generally (in this particularlinguistic/cultural context)A high-pitched is neither helpful nor harmful in response to a
householder low-pitched greetingHowever, a low-pitched interviewer greeting in response to a
high-pitched householder greeting is particularly detrimentalto participation
• Interviewers should be aware that as early as thegreeting, householders provide important informationregarding their specific expectations (cf. tailoring andleverage-saliency theory: Groves, Singer, & Corning 2000)
Practical ImplicationsWhy do some interviewers produce low-pitched hellos,
particularly following a high-pitched householderhello (125 contacts)?
• These interviewers may be less-skilled or less-experienced.
• Some interviewers may be anticipating a negativehouseholder reaction to a greeting implyingrecognition when in fact there is no recognition.
• Use of the prosodically-small greeting may alsoreflect interviewer recognition of the intrusiveness ofthe RDD call.
27
Next Steps• It is critical to consider the potential for interaction
between householders and interviewers in this line ofresearch.
• Based on positive findings on a controlled greetingtype, we can focus on specific linguistic hypothesesfor other greetings, such as “Hi” or “Good morning”
• We can also investigate other prosodic features ofthe greeting, such as duration and intensity
• Does householder greeting provide information of theeffect on participation of other features (i.e., after theinitial greeting) of the survey invitation?
• To what extent do particular interviewers vary in theiruse of householder greeting information? 28
Thank You
30
Sampling ContactsAll contacts from 5 studies
100 interviewers
40 or all cases per I withfirst contact agree/other
40 or all cases per I firstcontact refuse
All contacts from cases
Up to 80 cases per interviewer
positive negative
(27/100 Is worked on >1study)
31
Coding at the move level
32
Contacts from 5 Studies areIncluded:
• Gujarati (240)• National Study on Medical Decisions (358)• Interests of the General Public (56)• Mississippi Community Study (20)• Survey of Consumer Attitudes (706)• Total # contacts = 1380
Iwer turn duration in sec
• Agree: up to turn 13 8.95 (.25)
• Agree: turn 14+ 9.17 (.29)
• Refuse: up to turn 13 6.43 (.18)
• Refuse: turn 14+ 4.63 (.29)
• SCB: up to turn 13 6.89 (.14)
• SCB: turn 14+ 4.83 (.10)
34
Analysis
• At 6 levels:– Study: overall response rate, average length of
interview– Case: number of calls/ contacts, eventual outcome– Contact: day and time, length, disfluency rates,
disfluency convergence between interviewer andanswerer, overspeech, patterns of moves,outcome
– Interviewer: variance across contacts– Turn: linguistic measures– Move: disfluencies, type of move
35