Text > Talk > Code

Pragmatic aspects of the production of survey interview data(1)

by Paul ten Have©, University of Amsterdam


Interviews are a pervasive form-of-life in modern societies, but there are many kinds: news interviews, talk-show conversations, job interviews, medical consultations, etc. Standardized survey interviews, especially those taking place over the telephone, are a rather strange member of this already varied family. They are based on a rather strict script, specifying both questions and alternative answer possibilities. The interviewer is not the author of the questions he or she asks, but just has to read them as scripted, probe in a neutral fashion, and record the answers faithfully. This is a difficult task, but it is essential for the production-as-intended of survey data, on which much social science thinking and social policy is based.

There is a growing number of research reports on standardized survey interviewing, from a variety of perspectives. In the present paper, I will analyze a part of trajectory of the work of producing survey data in terms of the 'pragmatic contexts' which are involved, i.e. the projection of scripted questions, the readings of the questions, assisting in the production of the answers, fitting answers to scripted alternatives, coding answers, and any kind of 'repair work' that may be needed during the various parts of the trajectory.

The purpose of the paper is to specify the specific problematics of the work of survey interviewing as one of 'pragmatic context'. The idea is that in most kinds of interviews the 'pragmatic context' of any action in those interviews tends to be constituted by and in the flow of the 'local historicity' of the event. In standardized survey interviews this is most often not the case, at least not for the questions and the fixed answer alternatives. Therefore, the interviewer has to negotiate a reasonable compromise between the fixed elements and the local contingencies.

My analysis will be based on a consideration of a collection of empirical instances of actual tape recorded survey interviews, for which not only the interaction over the telephone is available, but also the questions to be asked (the 'script') and the codes, ultimately typed. These data therefore allow a detailed inspection in pragmatic terms of the sequence: Text > Talk > Code.

The research question, then, asks for the ways in which interviewers and interviewees in standardized survey interviews deal with the 'pragmatic contexts', both global and local, in which they do their work of questioning, answering and recordings answers.

The logic of standardized survey interviewing

'Standardized survey interviewing' serves different kinds of data production projects. On the one hand -- and this is the most simple and straightforward one -- it can be used to produce summarized answers that constitute the empirical basis of general statements about some state of 'the population' from which the set of respondents to this particular survey is considered to be a representative sample. An example would be 'Most people have trust in the current government', or more specific; 'x percent of the population says it trusts the current government'. A survey can also be used, however -- and this is the more sophisticated and 'scientific' project -- to produce evidence for general statements about the statistical relations between variables. A simple example would be: 'Older people have more trust in the current government than younger ones', or in a slightly more technical language: '"age" and "trust" are positively related', or still more sophisticated: 'the correlation between the variables "age" and "trust" is "+ .x"'.

In order to serve these purposes, the general requirement for the data gathering process, i.e. the questionnaire design and the actual 'standardized survey interviewing', is that the final results, i.e. the variation within the 'data matrix', is strongly related to the 'real' variation within the population, while the relation to variations in the data gathering process itself is weak. Standardization is one strategy to minimize the latter relation, by keeping the process 'constant' over the various instantiations, i.e. the separate interviews. The variation that does exist, should be 'unsystematic', that is, not related to the crucial variables, randomly distributed. In other words, the results should be 'valid', i.e. one should measure what one wants to measure, and 'reliable', i.e. independent of the particular properties of the instance of measuring.

The general purpose of standardization is to optimize these qualities, and especially to maximize 'reliability'. The variation between the answers recorded for different respondents should correspond maximally to the underlying differences between the people concerned, and other influences on the variation should be either minimized or randomized.

The general recommendations for 'standardized survey interviewing' are derived from this overall 'logic'. Formerly, surveys involved interviewers visiting respondents at home and working through a printed questionnaire, with in addition separate 'answer cards' that could be shown to respondent to choose from. Today, these interviews are mostly done using the technology of Computer Assisted Telephone Interviewing (CATI). This means that a questionnaire is implemented in a computer program and that interviewers have to call randomly selected telephone numbers and ask answerers for cooperation. The CATI-computer screen presents the interviewer with organized instructions for the interview, first, to read the displayed questionnaire sections to the respondent, and, second, to record the answers by choosing options from an available set of possibilities, by typing a corresponding code on the keyboard. The interviewers are instructed to follow the 'script' of the computer-implemented questionnaire very closely. They should read the questions as worded, probe in a neutral fashion, and not show any evaluation of the answers given. Interactional research of actual survey interviewing shows, however, that interviewers do not follow these instructions literally (Cf. Houtkoop, 1995, 1996, 1997a, 1997b; Maynard, Schaeffer, 1997; Schaeffer, 1991, with Maynard, 1996; Suchman &Jordan, 1990). Part of the analytic task, then, is to explicate the local logic implied in these 'departures'.

The pragmatic contexts of questionnaire design and instantiation

Although I don't have direct information on the specific purposes and contexts that played a role in the design process of the questionnaire at hand, some general properties of that pragmatic context can be inferred from the general logic explicated above and the concrete product embodied in the questionnaire items.

As an example, I will quote from the computer-based questionnaire the first 'real' question, i.e. following various introductory exchanges requesting cooperation and checking respondent eligibility given particular sampling strategies.

Data excerpt 1: Questionnaire item # 5 (line numbers added)

1 And now some questions about government agencies. As you
2 know, every 10 years there is a census of the population of
3 the United States. How confident are you that the Census
4 Bureau protects the privacy of personal information about
5 individuals and does not share it with other government
6 agencies - very confident, somewhat confident, not too
7 confident or not at all confident ?
14 ===>[goto 5b] [#goto added for p8003
The first sentence, 'And now some questions about government agencies.' (1), announces a series of questions, by mentioning their topic at a general level: 'about government agencies'. It thereby serves to restrict the area of relevance for the respondent. The next sentence, 'As you know, every 10 years there is a census of the population of the United States.' (1-3) offers a further restriction, the U.S. population census. So before the actual question is brought forward, its target area has already been indicated. The second of these is formulated as a kind of 'reminder', prefacing a supposedly commonly known fact with an 'As you know'. A questionnaire is formulated for 'any respondent', it is 'audience-designed' rather than 'recipient-designed' as are utterances in a conversation between mutually known participants (Houtkoop, 1995). For a general population survey, this means that any member of the population should be able to understand and process the questions. A 'reminder' of supposedly shared knowledge, not only specifies the area of interest, but also might help to create a situation in which people who do not know this common fact, might initiate a request for clarification.

These two sentences are followed by a third, which consists of two parts. The first of these is the actual question: 'How confident are you that the Census Bureau protects the privacy of personal information about individuals and does not share it with other government agencies'. This question is followed, in the second part of the sentence, by four alternative answers: 'very confident, somewhat confident, not too confident or not at all confident ?'. The question as a whole is therefore formatted as a forced-choice question. Note that the question itself is a double one, first referring to privacy protection, and then to not sharing information. As the second can be seen as a specification of the first, it may be seen as 'really one question'. As the excerpt shows, the interviewer is provided with four code options corresponding to the to-be-read alternatives, plus 2 'escape codes', which are not read aloud.

Now let us take a look at some realizations of this 'script' in actual practice, the interview.

Data Excerpt 2, Questionnaire item # 5, as in Interview 1(2)

80 IV: ·hhhh okay(gh): a::[::nd? now we have some questions=
81                        [##
82 IV: =about government agencies. ·hhh as you know:? every ten year
83     there is a census of the population of the united states. ·hhh
84     how confident are you: (.) that the census bureau protects the
85     privacy of personal information about individuals and does not
86     share it with other government agencies. ·hhh very confident
87     (0.4) somewhat confident (0.5) not too confident? (0.2) or
88     not at all confident. {q5}
89     (1.0)
90 FR: share it with what other governments
91 IV: (tch) ·hh well the question doesn't specify: but (0.3) it just
92     says other government agen[cie]s
93 FR:                           [oh ]
94 FR: probably very confident
95     0.5)
96 IV: °oh kay° people have different ideas about what the
We enter the interaction the moment the interviewer accepts a previous answer to a preliminary question with 'okay' (80) and enters it in the computer (81) -- so a few seconds before she starts to read the part of the script given in data excerpt 1, above. She immediately continues with the latter, announcing the first set of substantive questions 'about government agencies'(80, 82), and proceeds to ask the first of these (82-8). The respondent initiates a repair on one part of the question (90), which is dealt with by the interviewer (91-2) and accepted (93), followed by an answer to the question (94). After an accepting 'okay' the interviewer starts the next question in the series. So she reads her script rather faithfully, but in an unscripted rhythm and intonation. Then we see an inserted repair sequence, followed by a hedged scripted answer to the question, and an unscripted acceptance, which closes the sequence.

In an unpublished paper on 'Some design specifications for turns at talk in a job interview', Graham Button (1989) has coined concepts that are very useful to discuss multi-unit questioning turns like this survey 'question'. As he notes, the fact that many turns in a setting like the job interview are 'built out of multiple turn constructional units (...) contrasts with ordinary conversation where there is a pressure for the minimization of turn size'. He suggests that the general possibility for such complex, multi-TCU turns depends on a suspension of the pressure for the minimization of turn size that is operative in ordinary conversation. That pressure stems from the strictly local, turn-by-turn, organization of turn taking in conversation (Sacks, Schegloff, Jefferson, 1978). The suspension of this pressure depends on the participants' practical acceptance, that it is the interviewer who does the questioning. This means that when the interviewer, after an acceptance of the previously given answer, takes a turn which starts with objects like announcements, assertions, etc., anything other than a question, the expectation is that the turn is still incomplete. In examining his materials, Button has distinguished three different question components: a question delivery component (QDC), the unit which actually asks the question, and two others that may precede it: a question target component (QTC) and a personal relevance component. The Question Target Component is that part of the turn which 'is used to develop a target for the eventually delivered question', while the Personal Relevance Component 'displays that the relevance of the question resides in the candidate's expressed experiences'. It is only after the Question Delivery Component has been produced that the questioned can take the floor to work towards giving and answer(3).

Returning to the three-question components in our example, we can say that the assertions that precede the actual question, can be characterized as the Question Target Components. What I earlier called the actual question, corresponds to the Question Delivery Component, which in general tends to be the last in the turn, but is here followed by the list of answer options -- a phenomenon which is not evident in Button's materials. On the other hand, there is no Personal Relevance Component in these survey interviews, as they are 'audience-designed' rather than 'recipient-designed' (Houtkoop, 1995), as are job interview questions as well as conversational questions. So, the multi-component question format used here is essentially connected with the turn-taking system that is operative in these interviews, as it is in job interviews and news interviews. Because interviewers are expected to ask questions, especially as this activity has been announced as it is here, any component that is not recognizable as 'doing questioning', such as an assertion of supposedly common knowledge, is therefore taken as preliminary to an upcoming real question and therefore as a part of an as yet unfinished multi-component turn. In other settings where multi-component questions are regularly used, like news interviews as well as job interviews, they tend to be last in their turn. The fact that this is not the case here, as the answer option list still follows, is a special feature of survey interviews which may engender a specific type of 'trouble', as the respondent may be inclined to start answering the 'question' as soon as a Question Delivery Component is recognizably complete(4).

The general problematic of the situation, then, is that the interviewer has to 'enact', or rather, to use an expression inspired by Michael Lynch, 'bring to life' the generally formulated questionnaire in this particular situation, for this particular respondent. Part of this process is that the question reading gets a rhythm and an intonation contour(5). In the transcript as shown in data excerpt 2, we see indications of inhalations, prolongations, stresses, intonation changes and pauses, which, or course, were not available in the 'script' as given in data excerpt 1. Looking back at that excerpt 1, line 6, we can see that an effort seems to have been made to 'attach' the answer options to the question, at least typographically. This may have been done to suggest to the interviewer to produce an analogous 'attachment' intonationally, to ward off the 'trouble', noted above, of the respondent starting too early.

In the case at hand, we can see that the interviewer hearably inhales after the set announcement (line 82), and after the Question Target Component (83). This can serve as a kind of 'marking the structure' in a phase when that is still safe to do so, before the Question Delivery Component has been produced. That questioning component itself is produced in one breath and in a continuous rhythm, but it ends in a 'final' intonation, not marked as a question by a rising intonation. After an inhalation, the answer options are produced 'as a list', the items separated by pauses, the pre-final one with a rising intonation and the final item with a final intonation. This suggest an overall strategy of 'punctuation' to avoid a too early response. I will return to these issues in my later discussions.

When the question in data excerpt 2 is delivered, and the options list is brought to conclusion, there is a one-second pause before the respondent initiates a repair in the following terms: 'share it with what other governments' (90), that is, she repeats the target phrase with an inserted 'what', but leaving our the rather essential 'agencies'. The interviewer explains: 'well the question doesn't specify: but (0.3) it just says other government agencies' (91-2). After an Oh-receipt (93), the respondent answers the question in (94), which, although preceded by the qualifying 'probably', is accepted by the interviewer. By using a phrase like 'the question doesn't specify: but (0.3) it just says...', the interviewer takes a distance from the questionnaire and refuses to take the task of providing for such a specification on her own account(6).

Since this is a partially scripted kind of interaction, which is enacted again and again with different participants, it is interesting to compare the instance cited and discussed above with another one from the same set of interviews.

Data Excerpt 3, Questionnaire item # 5, as in Interview 2

61 IV: ·hh two a:nd now we have some questions about government agencies
62     ·hhh as you know every ten years (.) there's a census of
63     the population of the united states ·hhh how confident are you
64     that the census bureau protects the privacy of personal information
65     about individuals? ·hhh and doesn't share it with other
66     government agencies? ·hh are you very confident (.) somewhat
67     confident (.) not too confident (.) or not at all confident?

68     {q5}
69     (2.1)
70 MR: oh kay yer- you're talkin' (rapidly here you-)
71 IV: ·hh o[kay
72 MR:      [protect information from
73     (1.8)
74 MR: keep (things) confidential?
75     (0.7)
76 IV: (tch) well um ·hhh the:: question actually asks how confident
77     are you that the census bureau ·hh protects the privacy of
78     personal information about individuals and doesn't share it
79     with other government agencies {q5}
80     (2.0)
81 IV: so:: do you think the census bureau keeps thee information that
82     people give them? do you think they keep that private? and they
83     don't share it?
84     (2.0)
85 MR: eh:: i think they'd- they'd have to share it if
86     (0.6)
87 MR: gatherin' information
88 IV: ·hhh okay:? so:: how confident are you that (.) they:: don't
89     share it(h) huh ·hh
90     (0.8)
91 MR: uh::m not very con[fident]
92 IV:                   [·hhh  ] o[kay (.) A:ND people have=
93                                 [#
94 IV: =different ideas about what the census is ...
When we compare the way in which the 'same' question is produced in excerpts 2 and 3, we can see some interesting differences(7). In the second case, we observe the same addition we observed in the first case, 'we have' in the set announcement (here at line 61). And there is a further addition, 'are you', before the answer options (line 66). The spacing is rather similar, with inhalations and pauses at similar moments, except that the latter are shorter. The major differences are the inhalation between the two parts of the question (line 65) and the addition noted before. There are also differences in intonation, however, with rising ('questioning') contours at the end of both parts of the question and of the answer options. We might say that the specificity of this production resides in the rather fast tempo with which the items are produced, including the shorter pauses, and the repeated rising intonations on question parts and the optins, which seems to pack these as belonging to one 'whole', so to speak.

After this, we see a remarkably long pause, signaling an inability of the respondent to react (2.1 in line 69), for which he gives an account by complaining about fast speech (70) leading to a repair initiation which he enacts by repeating some key terms (72, 74). The interviewer then repeats the question (76-9), then paraphrases it (81-3), at which the respondent answers (85, 87). The interviewer, however, repeats the original question in truncated form (88-9), which the respondent answers in the required format (91); which is accepted (92) and entered in the computer (93).

Observe that like the previous case, the repetition of the question is introduced with a 'distancing' move: 'well um ·hhh the:: question actually asks' (76). When this strategy of repeating the question does not produce an immediate answer (a 2.0 silence in line 80), the interviewer uses a paraphrasing tactic (81-3), which leads, after another (2.0) silence (in 84), to a hesitant start of the answering (85, 87). We can observe here a typical order of repairing an understanding problem in these materials, first by a repeat, and when this does solve the problem, by a paraphrase.

When we consider the interviewer's uptake of the respondent's tentative answers (85, 87), we might say that in initiating a repair on these answers, she demonstrates that these are a possible 'trouble sources' for her. Her truncated repeat of the question (88-9), serves as a reminder of the requiredformat for an answer that in itself has already been accepted ('·hhh okay:?', 88). The 'so::' suggests that the respondent can produce the required formatted answer on the basis of her earlier spontaneous one. This episode, then, has a structure that can be modelled as: Respondent: tentative answer; Interviewer: provisional acceptance, plus format instruction; Respondent: formal answer; Interviewer: formal acceptance.

The fact that we see some trouble on this item in several interviews, may be related to the fact that it is the first 'real' question. Respondents need some experience to adapt to the actual interview routine, such as: the rather unspecific ways in which questions are worded; the way in which they are produced vocally; and the format requirements for answers that are contained in them.

In the collection from which these excerpts were taken, it can be observed that interviewers routinely produce an 'unscripted' vocal acceptance of the answer, mostly something like 'Okay' or 'mhm'. Furthermore, we may note that the audible keyboard clicks (marked as # in the transcripts) provide additional indications than the answer has been sufficient to have the interviewer to enter a code(8). By closing the previous question-answer sequence in this way, they also serve to open the floor for the next question-answer sequence, expected to be started by the interviewer. This expectation routinely is reinforced by the interviewer making vocal indications of starting a (longer) turn, by audible inbreath and 'u:m:'-like sounds. Furthermore, some of the questions start with a scripted 'And', while such an object can also be produced on the interviewer's own initiative (cf. Heritage & Sorjonen, 1994). So there seem to be a cluster of indications and devices to mark the end of one sequence and the start of a next one, all serving to pass the turn to the interviewer for a unit of speech that will have a question as its ultimate component.

It is against this background that I now want to consider two parts selected from later phases of the interview. I will first present the scripted text, and then discuss an actual realization in an interview (interview 1, the same as excerpt 2).

Data excerpt 4: Questionnaire item # 7d (line numbers added)

1 Sharing information between different agencies of government
2 when both want the same information saves time and money, but
3 it also means some loss in privacy for the individual. Do you
4 think the benefits of saving time and money outweigh the costs
5 in privacy, or do you think loss of privacy outweighs the
6 benefits of saving time and money ?


9 <8> DON'T KNOW
10 <9> REFUSED
11 ===>[goto 9]
In this part also, we have a an introductory statement, a Question Target Component, preceding the actual question, the Question Delivery Component. The first presents a dilemma, while the second asks the respondent to make a balanced choice. Remark that the QDC here includes the alternative options, which may be related to the fact that here there are two options, rather than four as in the previous example. From a designer's perspective, then, the amount of items one can combine in one 'unit' such as 'a question' is limited.

While this question may be seen to escape some of the difficulties of the one considered previously, it is apparently not without some problems of its own, as can be seen in the next excerpt which display it 'in action'.

Data Excerpt 5, Questionnaire item # 7d, as in Interview 1

147 IV: a:n' sharing information between different agencies of government
148     when both want the same information (0.4) saves time [( )
149 FR:                                                      [but
150     see now who's the different agencies of government.
151 IV: ·hhhh we:ll (0.4) the- you know the government (0.2) is made up
152     of different asiancies: one of which: is the census bureau
153     (0.9)
154 IV: so (0.2) they're just talking abou:t
155     (0.5)
156 IV: whatever: (0.4) the other government agencies are:
157 FR: within the united states
158 IV: >right<
159 FR: oh okay=
160 IV: =so w- y'know
161 FR: now i'm more clear
162 IV: ok(h)ay(h) huh huh ·hhh AH: sharing information between defferent
163     agencies of government [when both-]
164 FR:                        [i forgot  ]what was your question?
165    (0.3)
166 IV: Oh i'm- this is just the lead in
167    (0.3)
168 FR: oh o[kay
169 IV:     [we're just starting. ·hhh ah:: sharing information between
170     different agencies of government when both want the same
171     information (.) saves time and money? but it also mean some loss in
172     privacy for the individual. ·hhh do you think the benefits? of
173     saving time and money (0.4) outweigh the costs in privacy, ·hh
174     or: do you think the loss of privacy outweighs the benefits of
175     saving time and money {q7d}
176     (1.0)
177 FR: the first one
178 IV: okay benefits of saving time and money: outweigh:
179     (0.5)
180 IV: costs in privacy.
181 FR: °mm hmm°
182 IV: okay ·hhh you've given me your general views .
Here, the respondent already initiates a repair on an introductory statement (147-8), with a request for clarification (150). After a collaboratively produced explanation, the respondent produces a statement of satisfaction, closing this inserted repair sequence (159, 161). With a laughing 'Okay' and an inbreath, the interviewer starts reading the multi-unit question again, repeating the statement that has just been discussed (162). The respondent then interrupts this reading, saying 'I forgot what was your question?' (164). Whereupon the interviewer explains: 'Oh i'm- this is just the lead in' (166), which the respondent accepts (168), followed by another explanation: 'we're just starting.' (169). It is only then that the multi-unit question can be produced in full. In other words, the interruptive request for clarification has disrupted both the scheduled reading of the complex question and its hearability as such. It is only when the interactional table has been cleared, that the questioning interaction can return to normal flow.

The question reading here, as it was started in lines 147-8, and finally produced in full in lines 169-75, seems quite straightforward, without any additions and with largely expectable 'punctuations.

The final example I will discuss, see data extracts 6 for the script, and 7 for an actual realization, illustrated an actual case of the 'trouble' of early answering, suggested earlier.

Data excerpt 6: Questionnaire item # 9e (line numbers added)

1 Changes to our nation's health care system are now being
2 discussed by the President and Congress. A national health
3 care system could record some of the information now collected
4 by the census. Some people have suggested that this
5 information should be given to the Census Bureau for use in
6 the population census in order to reduce the cost of the
7 census and make the census simpler.
8 How would you feel about a national health system providing
9 information about you to the Census Bureau for use in the
10 population census ? Would you favor it strongly, favor it
11 somewhat, oppose it somewhat or oppose it strongly ?


16 <8> DON'T KNOW
18 ===>
In lines 1-7, the question target is explicated, followed by a Question Delivery in lines 8-10, which is quite general, as in 'feel about' (8). It is followed by a more focused question, i.e. 'favor', specifying four options to choose from. Observe that here the script has question marks for both component, while in the previous examples (excerpts 1 and 4), there was only one. Now let's look at how this may work in an actual case.

Data Excerpt 7, Questionnaire item # 9e, as in Interview 1

[first part of multi-unit question skipped]

206 IV: ·hh in order to reduce the cost of the census and make
207     the census simpler, ·hhh how would you feel: about a
208     national health system. providing information about
209     you to the census bureau for use in the population
210     census.
211     ·h[hh
212 FR:   [good idea
213 IV: (h)ok(h)ay(h) would you (.) favor it strongly (.)
214     favor it somewhat (.) oppose [it somewhat] or oppose=
215 FR:                              [ stro:ngly ]
216 IV: it stro:ngly
217 FR: strong[ly
218           [#
219 IV: °okay.° favor it strongly.
220 FR: mm hmm
221 IV: ·hhh would you be willing to give your social security ...
After the multi-unit introduction, the interviewer produces the first question component with rather strong stresses, but she ends it with a final intonation contour (lines 207-10). She then takes an inbreath (line 211), which seems to demonstrate that she is not finished yet. The respondent, however, already answers the question in her own words, 'good idea' (212). The interviewer laughingly accepts this, but continues by providing the options list as scripted (213, 214, 216). The respondent gives her response just after the relevant option is given, but before the option list is completed, and she gives it in truncated form: 'stro:ngly' (215). When the list is completed, she repeats it in the same format, 'strongly' (217), although in the mean time the oppose alternative has been mentioned using that same word (216). The interviewer enters the response in the computer (218), and subsequently verbally accepts is with a soft °okay.°, followed by the fully formatted favor it strongly.(219), which is acknowledged by the respondent. It is clear, then, that respondents do expect the QDC to be the last one, and that producing an options list after it leads to a disturbance of the interactional flow, requiring extra interactional work from both parties.

As to the format of answers, Suchman and Jordan observe that in survey interviews:

the interview schedule imposes external constraints on what form answers should take. These constraints result in somewhat contradictory problems: In some cases, responses that require elaboration are disallowed; in other cases, responses that in ordinary conversation are good enough in survey interviews require unreasonable elaboration.
Suchman & Jordan, 1990: 235

  The point is, of course, that the answer has to be given in a form identical with, or closely resembling, one of the answer options. Quite often, respondents produce material from which answers can be inferred rather than complete answers, or answers in another form, which for purposes of understanding seems quite adequate. In short, many answers confront the interviewers with an inferential coding task(9). Apparently, however, the interviewers are instructed to probe for formally adequate, rather than inferentially adequate, answers. This is what we see them doing in excerpts 3, 5, and 7. Such option-focused probes seem to instruct the respondents to limit themselves to formal answers, and abstain from any elaboration, demonstration, or explanation of their choices. At times, however, respondents do produce some indications of their personal reasons, but these tend to be disregarded, or acknowledged in a 'personal' fashion, with the interviewer stepping out of her professional role for a moment (Houtkoop, 1996, 1997b)(10).

Overall sequential organization and underlying epistemological structure

I will now consider the overall structural organization of these interviews, which can be characterized in the following way(11). After the opening a number of preparatory tasks of confirmation, sampling, permission to tape-record the interview, and assurances of confidentiality have to be fulfilled before the actual interview can start. Then there are a series of questioning chunks or 'modules', as one of the interviewers explains in a fragment not quoted here. The series contains both general attitudinal questions, relating to general issues, as quoted, and a series of personal question on respondents' characteristics, feelings, etc. The schedule ends with some factual questions for sampling purposes, and then the interview is closed off with further assurances and thanks. In this way, the interview is presented and enacted as the vocal filling of a form, consisting of a series of questions with fixed answer options. The order of the elements seems to be bureaucratically, rather than topically, motivated.

This overall format can be compared to that in other types of interviews, including medical interviews and semi-structured research interviews. In terms of its overall format, a medical interview has, as noted before, the 'naturally logical' form of a number of consecutive phases (Ten Have, 1989). The anchor point of this format is the complaint voiced by the patient on this or earlier occasions, which serves as a request for a service. It is this request which makes the ordinarily ensuing questioning by the physician relevant. The professional needs a clear conception of the expressed need, and additional information (in the medical case often involving physical examinations, tests, etc.), before the service (in the medical case: diagnosis and/or treatment) can be offered. Physicians tend to keep the agenda that presumably guides their questioning hidden from patients and they similarly tend to refrain from reacting to the information provided by patients, at least until they announce their overall assessment regarding the diagnosis and/or treatment. In short, physicians tend to maintain, and patients generally accept (at least openly, cf. Heath, 1992; Ten Have, 1995) an asymmetrical distribution of knowledge, lay, experiential, and personal on the patient side, professional and general for the physician. The overall structural organization, then, is related to what John Heritage (1997) has called: epistemological asymmetry(12). Various institutional arrangements reflect presuppositions regarding the distribution of knowledge, as well as cognitive abilities and interests.

In his research on semi-structured research interviews, Mazeland (1992, summarized in Mazeland & Ten Have, 1996) has developed an interesting distinction that can be related to these issues. In his materials he encountered rather marked differences in the overall structural organization of such interviews, or at least the organization of extended episodes. Some of these consisted of a fast turnover of small question-answer sequences, starting with relatively simply shaped, one-unit question turns. These he has called Turn-by-Turn (or TbT) Interviews. The contrastive type is distinguished by the fact that here the two speakers each take rather extended turns at talk, at least as 'primary speaker'. These he has called Discourse Unit (or DU) Interviews (cf. also Houtkoop & Mazeland, 1985). Such interviews (or episodes within interviews) started with a complex turn in which the interviewer gave some explanation of the sorts of information requested, often inviting a story-like format. Rather than specifying a relatively closed set of answer options, as in survey interviews, these were invitations to speak in the informant's own terms.

These two forms, Mazeland suggests, presuppose different distributions of knowledge. In the DU-format, the informant is constituted as a kind of expert on the knowledge sought by the interviewer, necessitating a kind of self-structured telling. Such formats are used with professionals like public prosecutors interviewed about their work organization, or interviews about extended personal experiences or exotic life styles. TbT-interviews, on the other hand, are used with topics which are quite general and non-specialized, including 'face-sheet' data, household organization, and schooling careers. In such areas, the cognitive structure of the topical field is apparently supposed to be so general that 'any member', including the researcher, has access to its structure, while each respondent can fill in the slots in the structure from his or her private stock of knowledge quite easily.

In general, researchers who use semi-structured interviews take great care to structure the interviews in a such a way that each topic seems to follow 'logically' from the preceding and 'naturally' leads to the next. One of the dilemmas of this format is that such a kind of topical organization implies a sharing of control, which may lead to rather irrelevant episodes or markedly forced shifts in topic. In DU-interviews, the interviewer initiates the topic, and then hands control over to the informant, restricting him- or herself to indirect control through selective reactions and probing questions. In TbT-interviews, such dilemmas surface in negotiations about the extendibility of the answers (Mazeland, 1992). Tensions between the cognitive interests of researchers and informant seem unavoidable (cf. Mazeland & Ten Have, 1996).

Returning now to standardized survey interviews, we can appreciate that the relatively arbitrary but strict topical organization of such interviews is related to different presuppositions regarding knowledge distributions and cognitive interests that are implicated in that 'form of life'. The respondents have no specific interest in some outcome of the interview, but seem to be motivated by a sense of civil duty. They sometimes acknowledge that they should have an interest in and opinion about the public issues about which they are questioned. The generality of the issues is quite often stressed in the introductory statements of the question or question series, such as 'As you know, every ten years there is a census of the population of the United States' (data excerpts 1, 2 & 3 above). In comparison with less structured forms of interviewing, a standardized survey interview generally has little to offer in terms of the pleasure of being able to express oneself on topics of one's own interest. Interview schedules, interviewers, and respondents seem to differ in the extent to which such expressive behaviour is allowed or restricted. The 'audience design' of the survey interview, then, is reflected on all the levels of organization we have considered.

Plans, practices, and accounts

Survey interviewing is done in the service of projects that are committed to ignore the kind of detailed local practices that are studied in ethnomethodology and conversation analysis. Survey research is planned on a high level of generality and its results are ultimately also processed on that level. It may be useful, in this context, to reflect a bit more on this 'difference' between the findings of EM and CA, on the one hand, and established conceptions of survey research, on the other. For this purpose, I'll make use of a proposal by Harold Garfinkel, in a text written with Lawrence Wieder (1992: 187-9), for some conventions to 'render' the findings of ethnomethodological studies, extracting this element from a more complex argument. They write (187): 'the locally produced, naturally accountable lived phenomenon of order* is referred to with ticked brackets--{order*}.' Furthermore, 'An arrow, , is used to refer to professional social analysts' skilled use of methodic procedures. Accounts, indicated by brackets.  i.e. (account), are specified by analysts with .' So we get something like: '{order*} (account), where '' is: 'the skilled use of methodic procedures'.

I would like to combine these notions with some ideas developed by Lucy Suchman in her Plans and situated action (1987). She discusses 'The planning model ... (which) treats a plan as a sequence of actions designed to accomplish some preconceived end' (28), and elaborates a contrasting view which stresses 'situated action'. 'The coherence of situated action is tied in essential ways not to individual predispositions or conventional rules but to local interactions contingent on the actor's particular circumstances' (27-8). In this view, plans never suffice as prescriptions for action. They have to be locally realized by changing the prescriptions in various ways. She suggests, however, that plans continue to be used to frame the actions, and that as such they are the basis for reports on and accounts for the actions, even if these have departed significantly from the plans.

Coming back to some of the 'findings' about survey interviewing, discussed before, we can use these notions and ideas to construct a kind of overall work-sequence for standardized interviews as a production site for survey data. The questionnaire, which I often called the script, can be linked to the notion of a plan. As I have demonstrated, in its realization as an actual interview, the plan is used as a framework, but it is also changed by being worked on in various ways -- through vocalization, reception work, and repair, for instance. In other words, the actual interview is a matter of situated (inter)action. The plan re-emerges, however, as the framework through which the respondent's reactions are (re)constituted as choices among the alternatives provided by the questionnaire. The answers are accounted for in the plan's terms. This suggests a double task of the interviewers: asking questions, that is, going down from the level of the plan to the level of the interaction, and noting answers, which amounts to bringing up the spoken reactions to the level of the plan again.

When we adapt Garfinkel's conventions to these findings, we get something like the following:

(questionnaire) {interview}(recorded answers)

or in more detail:

(questionnaire)                                                                                              (recorded answers)

{'reading' questions, understanding questions, giving answers, understanding answers, recording answers; and possibly 'repairing' questions and/or answers}

What this amounts to is that there is an essential gap between 'plans' (including instructions) and actions, and between actions and 'accounts' (including reports). Ethnomethodologists, and CA-researchers have an intrinsic analytic interest in the local practices of workers like interviewers and their interactional partners, the respondents. For survey researchers, however, such practices are 'essentially uninteresting', unless it can be demonstrated that these have systematic effect on the levels of generality at which such researchers reason. This contrast is also an issue of 'pragmatic context'.

* * *


1. Paper read at the 6th International Pragmatics Conference, Reims, France, 19-24 July 1998; Panel: Pragmatic aspects of standardized interviewing, organized by Hanneke Houtkoop-Steenstra & Charles Antaki. It is in part based on earlier explorations presented in the Workshop on Interaction in the Standardized Survey Interview, Free University Amsterdam, 18-21 November 1995, and at Essex'96, the Fourth International Social Science Methodology Conference, University of Essex, July 1 - 5 1996. I would like to thank the participants on those occasions for their helpful remarks, and especially Hanneke Houtkoop whose enthusiasm for the topic, as well as her published and unpublished writings has been a real source of inspiration. After the time of writing of this paper, most these writings have become available in a book (Houtkoop, 2000). The data, transcribed by Robert J. Moore, were provided through Nora Cate Schaeffer, University of Wisconsin-Madison. Parts of the current text have also been included in my book Doing conversation analysis: a practical guide. London: Sage, 1999.

2. The transcripts are based on the conventions generally used In conversation analysis, originally developed by Gail Jefferson. Some special conventions are: 'IV' stands for 'interviewer', while 'FR' indicates a female respondent, and 'MR' for male respondent. The ## signs, like those in line 81 indicate 'typing', the interviewer entering codes to represent a just-given answer. Brackets like in '{q5}' refer to questions in the questionnaire. Furthermore, I have printed those part of the spoken text that conform to the 'script' implemented in the CATI computer in bold.

3. Although I use Button's unpublished paper here, similar arguments have been put forward elsewhere, for instance in: Button, 1992; Greatbatch, 1988; Heritage & Greatbatch, 1991.

4. Cf. Houtkoop, 1995: 105, note 6.

5. In fact, there is only one change in wording, the additions of 'we have' in the set announcement (line 80); exactly the same addition can be observed in the second case, in data excerpt 3 below.

6. At the workshop mentioned before, Harrie Mazeland presented a number of instances of such 'distancing'. In terms of Erving Goffman's Frame analysis (1974: 517), the interviewer at such moments seems to excuse herself that she is not the originator of the words she speaks when she reads the script, that is, asks the question, she is only an emitter, and therefore not accountable for its wording and meaning. This aspect of survey interviewing is dealt with extensively in Hanneke Houtkoop's analyses, which were not yet published at the moment of writing, but have since then been published in Houtkoop (2000).

7. The interviewer is the same in both cases.

8. Cf. Greatbatch, Heath, Luff & Campion (1995), on the turn-taking relevance of keyboard work in a setting where interactional partners have visual access to the keyboard work of one of them, the physician in a consultation.

9. The idea that interviewers have a coding task has been elaborated by Tony Hak.

10. It seems probable that such elaborative accounts will be more frequent and extensive in face-to-face interviews, in the visual bodily presence of the interviewer, which Suchman & Jordan studied, compared with telephone interviews such as the ones studied here.

11. The formulation of an of 'overall structural organization' of a conversation was introduced by Schegloff & Sacks (1973) and has been taken up in a proposal by John Heritage (1997) for the use of CA to study institutional practices. He mentions 'six basic places to probe the "institutionality" of interaction': 1) Turn-taking organization; 2) Overall structural organization of the interaction; 3) Sequence organization; 4) Turn design; 5) Lexical choice; and 6) Epistemological and other forms of asymmetry.

12. Cf. note 11 above.


Button, G. (1989) 'Some design specifications for turns at talk in a job interview', unpublished paper

Button, G. (1992) 'Answers as interactional products: two sequential practices used in job interviews'. In: Drew, P., J. Heritage, eds. Talk at work: interaction in institutional settings. Cambridge: Cambridge University Press: 212-31

Garfinkel, H., D.L. Wieder (1992) 'Two incommensurable, asymmetrically alternate technologies of social analysis'. In: G. Watson, R.M. Seiler, eds. Text in context: studies in ethnomethodology. Newbury Park, etc. Sage: 175-206

Goffman, E. (1974) Frame analysis: an essay on the organization of experience. New York: Harper & Row

Greatbatch, D. (1988) 'A turn-taking system for British news interviews', Language in Society 17: 401-30

Greatbatch, D., C. Heath, P. Luff, P. Campion (1995) 'Conversation analysis: human-computer interaction and the general practice consultation'. In: A. Monk, N. Gilbert, eds. Perspectives on HCI: Diverse approaches. New York: Academic Press: 199-222

Have, P. ten (1989) 'The consultation as a genre'. In: B. Torode, ed. Text and Talk as Social Practice. Dordrecht/Providence, R.I.: Foris Publications: 115-35

Have, P. ten (1995) 'Disposal negotiations in general practice consultations'. In: A. Firth, ed. The discourse of negotiation: studies of language in the workplace. Oxford: Pergamon: 319-44

Heath, C. (1992) 'The delivery and reception of diagnosis in the general practice consultation'. In: P. Drew & J. Heritage, eds. Talk at Work. Cambridge: Cambridge University Press: 235-67

Heritage, J. (1997) 'Conversation analysis and institutional talk: analysing data'. In: D. Silverman, ed. Qualitative research: Theory, method and practice.London: Sage: 161-82

Heritage, J., D. Greatbatch (1991) 'On the institutional character of institutional talk: the case of news interviews'. In: D. Boden & D.H. Zimmerman, eds. Talk and social structure: studies in ethnomethodology and conversation analysis. Cambridge: Polity Press: 93-137

Heritage, J, M.L. Sorjonen (1994) 'Constituting and maintaining activities across sequences: And-prefacing as a feature of questioning design', Language in Society 23: 1-29

HoutkoopSteenstra, H. (1995) 'Meeting Both Ends: Standardization and Recipient Design in Telephone Survey Interviews'. In: P. ten Have, G. Psathas, eds., Situated order: Studies in the social organization of talk and embodied activities. Washington, D.C.: University Press of America: 91-106

Houtkoop-Steenstra, H. (1996) 'Probing behavior of interviewers in the standardized semi-open research interview', Quality & Quantity 30: 20530

Houtkoop-Steenstra, H. (1997a) 'Being friendly in survey interviews', Journal of Pramatics 28: 591-623

Houtkoop-Steenstra, H. (1997b) 'Tussen tekst en interactie; het gestandaardiseerde interview' [Between text and interaction; the standardized interview]' GRMMA/TTT - tijdschrift voor taalwetenschap 6

Houtkoop-Steenstra, H. (2000) Interaction and the standardized interview. The living questionnaire. Cambridge: Cambridge University Press

Houtkoop, H., H. Mazeland (1985) 'Turns and discourse units in everyday conversation', Journal of Pragmatics 9: 595-619

Maynard, D.W., N.C. Schaeffer (1996) 'Keeping the gate: declinations of the request to participate in a telephone survey interview', Sociological Methods & Research 26: 34-79

Mazeland, H. (1992) Vraag/antwoordsequenties. Amsterdam: Stichting Neerlandistiek VU (also: PhD dissertation University of Groningen)

Mazeland, H, P. ten Have (1996) 'Essential tensions in (semi)open research interviews'. In: I. Maso, F. Wester, eds. The deliberate dialogue: qualitative perspectives on the interview. Brussels: VUB University Press: 87-113 [also: http://www.pscw.uva.nl/emca/et.htm]

Sacks, H., E.A. Schegloff, G. Jefferson (1978) 'A simplest systematics for the organization of turn taking for conversation'. In: J.N. Schenkein, ed., Studies in the organization of conversational interaction. New York: Academic Press: 7-55 (1974)

Schaeffer, N.C. (1991) 'Conversation with a purpose--or conversation? Interaction in the standardized survey interview'. In: P. Biemer, et al, eds. Measurement errors in surveys. New York: Wiley: 367-91

Schaeffer, N.C., D.W. Maynard (1996) 'From paradigm to prototype and back again: interactive aspects of "cognitive processing" in standardized survey interbiews'. In: N. Schwarz, S. Sudman, eds. Answering questions: methodology for determining cognitive and communicative processes in survey research. San Francisco: Jossey-Bass: 65-88

Schegloff, E.A., H. Sacks (1973) 'Opening up closings', Semiotica 8: 289-327

Suchman, L. (1987) Plans and situated action: the problem of human-machine communication. Cambridge: Cambridge University Press

Suchman, L., B. Jordan (1990) 'Interactional troubles in face-to-face survey interviews', Journal of the American Statistical Association, 85: 232-41

* * *