Increasing Early Childhood Care and Development through Community Preschools in Cambodia 2016, Baseline Survey

Data management procedures for this study included:

• Testing of all instruments on Survey CTO v2.10. This included checking all of the items validity and logic patterns (skips, relevance rules, etc.). This stage also ensured that items used as key identifiers (for future merging of datasets) are clearly defined.

• Setting up on-field and in-office QC procedures (re-interviews, comparison of key items in different datasets, etc.). Implementing in-office procedures in order to identify and correct data inconsistencies and errors. Error reports were produced to this end.

• Setting up daily upload/download procedures and making sure that data is well received every day. Field work progress reports were produced to this end.

• Cleaning/labeling final dataset

Data validation and quality control took place during every step of the survey process, with mechanisms built into the survey design itself, data collection and data reporting processes. Thus, a range of tools was built into the data entry system to enable validation of data both at point of entry and during reporting stage. Unique identifiers were developed for each questionnaire in advance, to assist with datasets merging and longitudinal tracking.

Quality control procedures: During data collection, frequent checks of the data entry system were undertaken, as well as ongoing comparison of re-interviews, to ensure that data is accurately recorded by interviewers and the system is functioning as designed. After completion of data collection, the complete database was “cleaned” to check for inconsistent or missing values, incorrect skips, and validation errors. In addition to the in-system validation rules mentioned here before, the following QC procedures were implemented:

• Interview Summary: A summary of critically important items was displayed as in-system feature at the end of the interview. These items were mostly used for the dataset merging and needed to be exactly identical between the household/caregiver and the child systems. Interviewers and Field Editors made sure that the same merging information was recorded for both systems.

• Interview observations: The Field Editor or the Field Supervisor randomly selected an Interviewer in their team and joined the interview at any random point. They observed whether the Interviewer actually implemented the interview techniques agreed upon during the training: respect of the question exact and accurate wording, following conversational or standardized techniques on the relevant items (defining or not difficult notions/words to help the respondent answering the question), accurately recording respondent answers into the tablet system (Survey CTO v2.10), following ethical policies, etc.

• Re-interviews: The Field Editor randomly selected a household, from those with who an interview was completed. He/she then re-interviewed this household, asking once again a selection of 15-20 items (chosen because of their critical importance: items used for merging, items used to assess household eligibility, filter questions, etc.) for a maximum of five minutes. Stored in a separate dataset, this data was then compared against the actual initial interview data, back in office, by the Data QC Supervisor. Discrepancies were inquired into and the Data QC Supervisor called the respondent back to obtain the final correct answer. Observations and re-interviews concerned a minimum of 20% of the cases (around 5 households per village).

• Time stamps: In-system section time stamps were analyzed. They allowed the Data QC Supervisor and the Data Manager to identify potential issues (if a section/interview took seemingly significantly more time than the average) and to help assessing fieldwork progress and forecast fieldwork completion.

At completion of the data collection, (potential) data entry of paper forms and reconciliation processes, a draft database was produced, accompanied by a codebook and will be cleaned and labelled using Stata 13.1 do-files. The database was finalized based on feedback from the WB research team.

Qualitative data (as necessary: Concurrent with the quantitative data processing, any qualitative data (open-ended data from household and individual interviews) was translated into English by experienced Khmer-English translators.