We measured momentary well-being using the Experience Sampling Method (ESM) among 220 water collectors in rural Meru County, Kenya over eight weeks. Subjects reported on affect and time use at four randomly-chosen times through the day (Monday through Saturday) on a custom-designed ODK survey app, deployed on a low-cost smartphone. Subjects completed a second ODK survey each weekday evening, reporting on school attendance, study time and chores performed for each school-aged child in the household. After several weeks of baseline data, half of households were randomly chosen to receive free delivery of water to their door for four weeks, reducing water collection times to (near) zero. In-person baseline, midline and endline surveys were conducted by enumerators. The dataset “Meru ESM RCT.dta” contains (in Stata format) the merged data from the ESM exercise and the baseline, midline and endline surveys. The baseline, midline and endline survey were conducted once with each household, but each household completed multiple ESM surveys. This dataset contains 12,956 observations, so to recreate the baseline, midline and endline datasets (one row per household) one would collapse the data on phoneid. The baseline, midline and endline surveys contain some data and questions that were repeated across waves. To make variable names unique, a “_base”, “_mid” or “_end” is appended at the end of the variable name. For example, each survey contained the time that the interviewer opened the app and started the survey (start), as did the ESM survey completed by the subject. This dataset therefore contains four variables, start (the ESM surveys), start_base, start_mid, and start_end. All data was collected in ODK apps. These apps are compiled based on data in Excel spreadsheets, including variable names, questions, and answer codes. These ODK excel spreadsheets thus also serve as data dictionaries. The key unique identifier linking records is phoneid. This is an identifier created by the team, and is not a phone number, or SIM serial number that is any way identifiable. With few exceptions, data files do not contain any variables generated ex-post by the researchers. The exception is a variable tracking treatment status. The original, randomly-assigned treatment status is encoded in treat. But because implementation of the treatment program was uneven in the first week (and particularly the first two days) due to logistical issues, the team created three time-varying treatment variables capturing three assumptions. trtsimple drops all treated households from the dataset during the first two days of treatment. trtconserv drops all treated households until the water delivery system was running as planned, dropping approximately 2 weeks of data. trtmain uses detailed information collected by the study team about the dates and locations where water delivery was operating as expected. In other words, if the team has data that treated households in a given location were delivered water as planned on a given day, we do not drop those treated households even if other treated households elsewhere did not receive water as planned.
See papers for more details.
Probability
Sannolikhetsurval
Face-to-face interview: CAPI/CAMI
Personlig intervju: CAPI/CAMI
