Page 391 - Special Topic Session (STS) - Volume 4
P. 391
STS2319 Lakshman N. R. et al.
where, in this case is the total number of errors (sum of skip, logical,
validation, and missing) committed in the questionnaire of household i
interviewed by enumerator j in cluster c. Again, = 1 if the mode of
collection was CAPI and 0 if otherwise. and are the household
characteristics and enumerator characteristics, respectively. The fixed effects
employed for the model include enumeration area and enumerator code.
Finally, to assess how the costs of CAPI stack up against traditional modes,
back of the envelope calculations were done looking at fixed and variable
costs for CAPI and PAPI to determine the break-even point. The simple
arithmetic for such is as follows:
+ ( × . )
= + ( × . )
where FC is the fixed cost and VC is the variable cost to conduct the survey.
3. Results
The result of the first model investigating CAPI’s effect on interview time
is summarized in Table 2. The model strongly suggests that CAPI reduces
interview times by about 28.5 minutes. This finding is significant and robust
even when accounting for enumerator and household characteristics.
The model also finds that across both modes it seems that females’
enumerators tend to conduct shorter interviews. Meanwhile, it is interesting
to note that enumerators that possess higher education (college) seem to be
associated with shorter interviews. Duration seems to decrease with age, while
more enumerator experience seems to be linked to longer interviews. A
possible reason for this could be that more experienced enumerators might
just be more meticulous and detail oriented in administering the questions
and probing for answers. Also, as expected, the number of adult household
members contributes to the duration of the interview positively.
The results of the second model are presented in Table 3. Our study finds
that errors are statistically significantly reduced with CAPI (Table 3). To put into
perspective, moving from PAPI to CAPI is associated with reducing the error
incidence by about 1.5 per questionnaire. Our study also suggests that
regardless of whether CAPI or PAPI is used, household characteristics still do
matter as far as the reduction of errors, while only the gender of the
enumerator has implications for data quality. Most importantly, using CAPI
does not mean that errors are completely random, thereby producing
unbiased statistics. This has implications for data analysis and policy.
380 | I S I W S C 2 0 1 9