Optional Subject: Field Data with a Convenient Gap
A graduate student in an optional-subject research methods course is analysing household survey data for a district study on remittance use. The dataset has 480 households, but 61 records are missing the income variable. A regression excluding those records produces a strong, publishable result; a version using mean imputation produces a weaker, non-significant result. The student's supervisor is travelling and has asked for a draft chapter by Friday. A classmate suggests dropping the 61 records and 'not dwelling on it' in the methods section. The student notices that the missing records cluster in one sub-district that is known to be poorer, and the survey enumerator's notes mention that respondents there were reluctant to disclose income.
1. Assess whether listwise deletion is defensible here, and explain the statistical and ethical issues raised by the clustering of missing data.
Read the worked answer
Listwise deletion assumes data are missing completely at random, but the clustering in a poorer sub-district with documented reluctance to disclose income suggests the missingness is related to the variable itself. Deleting those records therefore biases the sample toward wealthier, more forthcoming households and inflates the apparent strength of the result. The 'strong, publishable' finding may be an artefact of who was excluded.
Ethically, omitting this from the methods section would misrepresent the analysis and could mislead policy readers about a vulnerable sub-district. The classmate's suggestion to 'not dwell on it' is a disclosure failure, not a stylistic choice. The student should report the missingness pattern, justify the chosen method, and present sensitivity analyses rather than selecting the result that pleases.
2. Recommend an analysis and reporting approach the student can defend to the supervisor, including how to handle the Friday deadline.
Read the worked answer
Report the missingness pattern explicitly by sub-district, then use a method appropriate to non-random missingness - for example multiple imputation with auxiliary variables, or a sensitivity analysis comparing listwise deletion, mean imputation, and imputation under plausible assumptions about the poorer sub-district.
Present the primary result alongside the sensitivity results so readers see how conclusions shift. Avoid selecting the specification that yields significance.
On the deadline: submit the draft with the sensitivity analysis and a short memo to the supervisor explaining the missingness issue and requesting guidance before finalising. Meeting Friday with a transparent, provisional analysis is preferable to meeting it with a concealed methodological choice that could unravel at review.