Jan 2027
18 Mon
19 Tue
20 Wed
21 Thu
22 Fri 09:00 AM – 06:00 PM IST
23 Sat 09:00 AM – 06:00 PM IST
24 Sun
Kartik
@kwatrakartik
Submitted Oct 25, 2026
We often analyse data that is already generated, but the data generating process plays a huge role in how reliable the conclusions from the data are. This talk is about the issues in conducting surveys (a data-generating process) and some of the ways you can pre-empt them or address them post facto.
Every stage of a survey, from questionnaire design and sampling to fieldwork, shapes data quality, which shows up in two ways: the precision of the estimates and their bias. I will focus on fieldwork, where many problems arise that the design never accounts for, and that are easily missed unless you look for them. Some of these can be spotted in paradata (data about the process of collecting each response): how the questionnaire was canvassed (e.g. the flow of questions), the time of day of interviews, interview durations, location patterns, interviewer and team-level patterns, and deviations from the prescribed interviewing methodology. Others show up in the response data itself, such as suspiciously low variation within an interviewer’s work, answers that contradict each other, or age heaping (ages clustering on round numbers). For each, I’ll show how to spot it and what,if anything, can be done about it, both during fieldwork and after. The methods range from statistical measures of reliability and consistency to patterns in paradata and responses that I’ve learned to watch for in my survey work.
I will be using synthetic data that mimics issues I’ve found in real fieldwork to illustrate the points.
Takeaways:
For practitioners commissioning surveys:
For consumers of survey data:
About me:
Kartik Kwatra
Data Scientist
I’ve been doing end-to-end data work at How India Lives for over 8 years, with a focus on public data. I have worked on pan-India surveys, rural and urban, across every stage: questionnaire design, sampling, overseeing fieldwork, and analysis and reporting.
Linkedin
I’d like to meet: researchers / practitioners / domain experts
Topic/domain tags -
data quality, survey data, paradata
Availability before the event: yes
Hosted by
{{ gettext('Login to leave a comment') }}
{{ gettext('Post a comment…') }}{{ errorMsg }}
{{ gettext('No comments posted yet') }}