Zambian Financial Diaries Project 2015

Transactions data:
Date start and date end variables are fuzzy (there are not always seven days in a week, the weeks don't always begin on the same day) which is most likely attributable to data capturing errors on the part of the fieldworker. Most of the week lengths, when evaluated, come to seven days (as expected) but not all. For the user the more reliable measure of the week of observation is the transact_week variable.

Roster data:
The wards variable seems to be imperfectly captured as many do not match lists of recorded Zambian electoral wards. Efforts have been made to make the entries more readable but are imperfect perfect. The phone access variable is also bit misleading. There are 58 missing values for the variable roster_phoneaccess_or_own which seem to have a corresponding follow-up response with variable roster_accessph_only. It is unclear what the roster_accessph_only} variable is meant to represent because it is not on the enrolment questionnaire.

Cross-section data:
There are two sets of duplicates in terms of RespID in the cross-sectional data file. Investigating further it was discovered that one set of duplicates is the apparent result of the same fieldworker visiting the same household twice (the second visit occurring eight days after the first). This policy adopted here was to choose the latest observed row in the data. Notably, there were a few variables that had different values between the two observations. These are easily attributable to actual dynamics within the household. There were no substantive differences between static household characteristics.

The other set of duplicates was slightly more complicated. It involved the apparent incorrect assignment of the RespID code (that is, the RespID code was assigned correctly for one entry and incorrectly for the other). Fortunately, it was possible to check the correct RespID using the other data files. It turned out that the incorrectly assigned entry was meant to be represented by another code entirely. This is most likely a data capturing error. Correcting the incorrectly coded RespID yielded another set of two duplicates. The correctly coded entry of these two duplicates was the one used in the final file.