In the last segment we talked about interviewer effects that are reflected by

increased interviewer variance and

we saw that this seemed to be due to interviewer behavior.

Such as when and how effectively interviewers probed,

the one aspect of standardized interviewing that's not standardized.

We now turn to the other type of interviewer effect in the methodology

literature, namely differences in response distributions due to interviewers'

enduring or fixed characteristics such as race and gender.

A well-known study of gender of interviewer was done by Kane and Macaulay.

And they found that both male and

female respondents gave more egalitarian answers to female than male interviewers.

Egalitarian answers to questions that concerned gender somehow.

But, that the male and

female respondents showed to effect on different survey items.

In particular more men reporter sharing childcare when the interviewer was female.

More women supported feminist policy when the interviewer was female.

There was no impact of interviewer gender when the topic of the question

was not relevent to gender.

So this affect showed up just for questions that concerned gender somehow.

Another study by Huddy and colleagues report a number of

gender of interviewer effects in two pre-election surveys.

In this case, male and

female respondents showed the same gender of interviewer effects.

As you can see those here, in this table, questions that produce

gender of interviewer effects namely, differences between answers to female and

male respondents on a ten point scale included feminist identity,

egalitarian roles, such as childcare and who does the house work.

Pro-choice, pro-affirmative action, among others.

So it can make quite a difference whether the interviewer is male or female for

questions, again for which gender is relevent.

Race of interviewer effects are analogous to gender of interview effects in

that they're only observed for questions for which race is relevent, and they're

observed typically for face-to-face interviews, although race of interviewer

has been shown to affect responses to questions about race over the telephone.

These results from Hatchett and

Schuman illustrate what we mean by race of interviewer effects.

In this case the respondents were white.

As you can see when the interviewers were black, more respondents

endorsed the particular position than when the interviewers where white.

So for example, more respondents said they would not mind if a relative of theirs

married a negro when the interviewer who asked the question was black,

than when the interviewer was white.

Or more of these white respondents Indicated they believe negro and

white students should go to te same school.

Substantially more respondents indicated they would not be disturbed if a negro of

the same class moved into the block they lived in when the interviewer who asked

this question was black than when the interviewer who asked it was white.

And more respondents indicated they believed negro and

white children should play together freely when the interviewer

who asked the question was black than when the interviewer was white.

A similar pattern is shown in a study by Schuman and

Converse for black respondents.

More of these respondents indicated they could trust white people,

trust most whites when they interviewer who ask a question was white than

when the interviewer who asked it was black.

Or more of these respondents said yes when ask if negro parents

work best with a negro teacher.

Fewer respondents said yes when the interviewer who asked the question was

white than black and

fewer respondents named only black entertainers when asked who their favorite

entertainer was, if the interviewer who ask the question was white, then black.

Now the first two of these

effects are probably due to a kind of social desirability.

A desire to answer the question in a way that these respondents believe would

be less offensive to or would please the interviewer given the interviewer's race.

The third of these, is the kind of effect that's often attributed to kind

of a priming or activation of stereotypes and beliefs.

So just seeing a black interviewer's face

may bring to mind black entertainers, more than single white interviewers face.

It could be that this is not about social desirability at all, this third effect,

but really just the examples of entertainers that

are brought to mind by the attributes of the interviewer who asked the question.

So teasing apart these effects that are due to social desirability and

due to kind of priming or activation, is complicated.

But some progress is made in a study by Krysan and

Couper in which they had interviewers ask racially relevant questions

either in face-to-face interviews or in video-recorded interviews.

So they were live and recorded.

They called the video-recorded interviews video ACASI, so

this allows one to compare social desirability in live interviews to

stereotyped activation in video recorded interviews.

The pattern of the results they report is complex, but it does help

us distinguish between the two origins of this kind of interviewer effect.

For example, they reported that African American respondents reported less liberal

racial attitudes to white than black interviewers in live interviews.

As we would expect, consistent with the two studies we just talked about, but

also in recorded interviews.

So, the first of these effects is probably a social desirability effect.

The second of these really can't be social desirability,

because the interviewers are recorded and the respondents wouldn't be

motivated to answer in a way that pleases the recorded interviewers.

But it could be much like we saw with

the virtual computer animated interviewers in our discussion of audio CASI.

It could be that the social presence of the recorded

interviewers triggers the same kinds of reactions that a live interviewer does.

White respondents reported

more conservative racial attitudes to black than white recorded interviewers.

This, too, is attributed to stereotype activation.

But the pattern is in the opposite direction.

So if the pattern were in the same direction as that

shown by the African-American respondents,

then these white respondents would have reported less

conservative racial attitudes to black than white recorded interviewers.

But the author suggests that this is a case in which the black

recorded interviewer's face served to prime existing attitudes about

African Americans which these white respondents truthfully,

faithfully reported, presumably because they were not inhibited.

They were not inclined to present themselves in a favorable light

to a video recorded interviewer.

So theses patterns may seem or the interpretation of these patterns may seem

somewhat ad hoc, but really if you step up a level they all demonstrate,

all these results demonstrate that the respondents are highly sensitive to

racial attributes of interviewers, whether live or recorded.

And this study helps distinguish between those different, the different origins for

this kind of effect.

Now we focused almost entirely on face-to-face or

at least interviews in which facial features are visible.

Then we recorded, video recorded interviews.

What about telephone interviews?

Is it possible that there are race of interviewer effects on the telephone?

The answer is yes.

Just one example comes from a study of Davis,

who examined race of interviewer effects in the 1984 National Black Election Study.

This was a telephone survey of African-American respondents.

There were 48 white interviewers and 27 African-American interviewers.

The effect that demonstrates a race of interviewer effect, the finding,

which demonstrates a race of interviewer effect

Is a little bit of a kind of second order effect.

It's a correlation between the answers to two questions.

A question about racial consciousness, and support for Jesse Jackson,

who was the first viable African-American candidate for

the Democratic Party's nomination for President in the 1980s.

And the finding was that the correlation between these two items was higher,

not high, but higher when the interviewer was African American than when

the interviewer was white.

So the idea is that a respondent would be more likely or

more willing to report being both a high in racial consciousness,

an African-American respondent report being high in racial consciousness and

at the same time supporting Jessie Jackson.

It's as if with white interviewers these African-American respondents were

moderating the impression they gave the interviewers.

Again this is on the telephone where the only information about the interviewers

ethnicity or race is communicated through voice.

So one study that bears on the issue of how accurately do respondents

classify the race of the interviewers who are speaking to them on the phone,

was done by Davis and Silver.

And they found first of all,

that respondents are actually can be quite inaccurate,

their ability to judge race varied between 14%, quite poor, and 82%, quite good.

The substantive finding was that black respondents

correctly answered fewer knowledge questions, political knowledge questions

when they perceived the interviewer asking the questions on the phone

to be white than when they perceived the interviewer to be black.

The authors attribute this to what's known as stereotype threat which is the tendency

for minority students to perform worse on standardized tests than white students,

because they're exerting so much effort and investing so

much attention in disconfirming the negative stereotypes that they believe

others ascribe to them, mainly that they'll perform poorly.

They work hard to disconfirm that and as a result a are distracted from the task and

end up a performing poorly and the authors argue that knowledge questions

which are a relatively common on surveys particularly political

surveys are very close to standardize test and so could trigger a stereotype threat.

So the point is the black response perform worse when they perceives the interviewers

as white And this was determined through questions by the researchers about

what race do you think the interviewer is, but when the authors took in account

the interviewer's self reported race this effect, this reduced performance

when black respondents thought Of the interviewers who were white went away.

And so this suggests that while respondents may find it

hard to not classify interviewers' race on the basis of what's on the phone,

they're not very good at it.

And so they may react accordingly, but they're actually

incorrect in the judgments that are triggering these behaviors.

One other type of effect that can involve race is,

we've already talked about, and it's social distance in which that's defined

as the match between attributes, the attributes of two people.

So in this case the attributes between the interviewer and respondents.

As you'll recall what Johnson and his colleagues found was

that the respondents were more likely to report drug use

the more attributes they shared with the interviewer including race and ethnicity.

This study too is conducted on the telephone.

This is just another indication that respondents

are sensitive to race as well as other enduring attributes of the interviewers.

Johnson and his colleagues showed that

the more attributes the interviewers and respondents share with each other,

the more drug use respondents reported both lifetime and recently.

And so, these attributes include race and ethnicities.

And it's just one more indication that tespondents even on the telephone

are quite sensitive to the attributes that they at least believe

are communicated through the voice of speech of the interviewer.

So to summarize our discussion of interviewer

effects that are due to fixed attributes, such as race and gender,

these attributes matter to the extent that the questions concern those attributes.

And when they do, answers can be quite biased.

So for example, more respondents reported being pro-choice in

the questions about abortion, when asked by female than male interviewers.

And these kinds of effects are stronger in face-to-face, than telephone interviews,

but they are observed in telephone interviews,

at least, they are observed to the extent that respondents' perceptions

of the interviewers' attributes are accurate.

As we just reviewed social distance based on demographic overlap

can affect disclosure also when communicated over the phone.

The more overlap the more disclosure.

So, respondents really do detect these and

attend at some level to these attributes when formulating their answers.

Beyond training interviewers to administer questionnaires as professionally as

possible, there's probably little one can do to reduce these kinds of interviewer

effects at the source, although they can be addressed analytically.

So to summarize our entire discussion of interviews and interviewers,

we first talked about interviewers in the role as recruiting units,

obtaining interviews and also selecting respondents from within the household.

Interviewers ability to obtain interviews can be affected by how they interact with

sample members, and what their vocal attributes are like.

With respect to selecting respondents within the household, we discussed the way

particular selection procedures lead to more and less representative samples.

We also discussed interviewing technique and found that this can have

a large affect on the quality of answers,in particular response accuracy.

And it may do this without increasing or affecting interviewer variance.

We spent considerable time looking at the comparisons between conversational and

standardized interviews where we saw that conversational interviewing can

substantially increase response accuracy.

But the tradeoff is that this increased accuracy takes more time.

The interviews are longer.

It takes more time to collect more accurate data.

Interviewer variance is about the same in conversational interviews as it is in

standardized interviews.

And in the last two segments, we've talked about interviewer effects,

both due to interviewer behavior and interviewer attributes.

The main point is that interviewer effects can add error to the survey estimates.

But overall, interviewers continue to

add value to the survey enterprise by increasing response rates,

clarifying question meaning and motivating respondents and keeping them on task.

In our next and final lesson, we'll turn to new modes of data collection such as

mobile web and SMS, and also examine the potential for using

new data sources such as social media in place of conventional survey responses.

3.3.2 Bias: Interviewers’ Fixed Attributes

Data Collection: Online, Telephone and Face-to-face

与讲师见面