Do Survey Spot-Check and Threat Improve Data Quality? Evidence from a Field Experiment

Abstract

In this study, we examine the effect of interview observation on data quality of a large-scale individual survey in Vietnam. External spot-check teams were sent to randomly selected enumeration areas to attend and observe survey training courses and field interviews of interviewers. We find that interviewers have a lower number of item non-responses in collected data when being observed by the spot-check teams. The effect of a spot-check of field interviews appears to be higher than the effect of a spot-check of training courses. However, we do not find a significant effect of a spot-check threat: informing interviewers of a possible spot-check without actually visiting does not reduce item non-responses.

Acknowledgements

The experiment in this study is funded by the United Nations Development Programme (UNDP) in Vietnam. We are very grateful to Do Thanh Huyen, Sarah Dix, and Jairo Acuña-Alfaro, Edmund Malesky, and two anonymous reviewers of the Journal of Development Studies for their very helpful comments on this study.

Disclosure statement

No potential conflict of interest was reported by the author.

Supplemental Material

Supplementary Materials are available for this article which can be accessed via the online version of this journal available at https://doi.org/10.1080/00220388.2018.1539474

Notes

1. Non-sampling errors can be classified into non-response errors and measurement errors (Banda, 2008). Non-response errors occur when a survey fails to obtain the intended information from the respondents. Non-response consists of both unit non-response and item non-response. Unit non-response implies that no information is obtained from certain sample units because the respondents refuse to participate in the survey or cannot be contacted. Item non-response refers to a situation where the information collected for some units is incomplete.

2. The 2014 PAPI survey is the most recent survey at the time of writing this paper.

3. The regression results are reported in Tables S.8 and S.9 in the Supplementary Materials.

4. This is done in Stata using command ‘ivreg2’.

5. PAPI 2009 and 2010 are just pilot surveys which did not cover all 63 provinces. Thus we do not use these datasets in this test.

Additional information

Funding

This work was supported by the United Nations Development Programme (UNDP) in Vietnam [2014.04].