Inflated Sampling Frame and Substituted Respondents in a UPI Adoption Survey

Two separate biases are present. First, the frame is inflated: 210 of 1,450 entries are ineligible, so the eligible frame is at most 1,240 and the true response rate is 456/1,240 ≈ 36.8%, not 38% of a clean frame. Coverage error also arises because the 44 nearby-market vendors were outside the defined population; their adoption behaviour may differ systematically. Second, non-response is not random: 62 vendors were never contacted and 44 were replaced, with no refusal reasons recorded. Vendors who are less digitally active may be more likely to be absent from the market or to refuse, so the 71% figure is plausibly upward-biased. The municipal benchmark of 44% is consistent with this direction, though it measures transactions rather than self-reported adoption. The 44 substituted cases should not be pooled with the probability sample; if retained, they must be analysed separately as a convenience supplement.

The scholar should first reconstruct the eligible frame by removing duplicates and closed stalls and by documenting the rule used. The response rate should be recomputed against the eligible frame and reported with the number of refusals, non-contacts and ineligibles. A short non-response follow-up should be attempted, at least by phone, to compare early and late respondents on key variables. If reliable vendor counts by section or vendor type are available, post-stratification weights can reduce non-response bias; otherwise a sensitivity analysis should show how the estimate changes under plausible assumptions about non-respondents. The 44 substituted vendors should be excluded from the main estimate, or reported only as a separate convenience sample. The final report should present the corrected estimate with a confidence interval and explicitly compare it with the municipal transaction benchmark rather than presenting 71% as a population estimate.

The ethical problem is transparency and informed consent, not merely statistical error. Respondents from the nearby market consented to a survey, but they did not consent to being presented as part of the original market sample. Silently replacing non-respondents also creates a misleading impression of coverage and can amount to misrepresentation of the sampling method. The scholar should disclose the replacement in the methods section, explain why it occurred, and remove those cases from the population estimate. If the data are used at all, they should be labelled as a supplemental convenience sample. The scholar should also check whether institutional ethics approval covered the changed procedure and should retain consent records. The guide should require a limitations paragraph that states the likely direction of bias, rather than allowing the 71% figure to stand unqualified.