Household Registration Study 2015

Managing and Cleaning the Data

Data were managed and cleaned each day immediately upon being received, which occurred at the same time as the fieldwork surveys. At the end of each workday, the survey teams were required to review all of the interviews conducted and transfer collected data to the server. The data received by the main server were downloaded and monitored by MDRI staff.

At this stage, MDRI assigned a technical team to work on the data. First, the team listened to interview records and used an application to detect enumerators’ errors. In this way, MDRI quickly identified and corrected the mistakes of the interviewers. Then the technical team proceeded with data cleaning by questionnaire, based on the following quantity and quality checking criteria.

• Quantity checking criteria: The number of questionnaires must be matched with the completed interviews and the questionnaires assigned to each individual in the field. According to the plan, each survey team conducted 20 household questionnaires in each village. All questionnaires were checked to ensure that they contained all essential information, and duplicated entries were eliminated.
• Quality checking criteria: Our staff performed a thorough examination of the practicality and logic of the data. If there was any suspicious or inconsistent information, the data management team re – listened to the records or contacted the respondents and survey teams for clarification via phone call. Necessary revisions would then be made.

Data cleaning was implemented by the following stages:

1. Identification of illogical values;
2. Software – based detection of errors for clarification and revision;
3. Information re-checking with respondents and/or enumerators via phone or through looking at the records;
4. Development and implementation of errors correction algorithms;
   The list of detected and adjusted errors is attached in Annex 6.

Outlier detection methods
The data team applied a popular non - parametric method for outlier detection, which can be done with the following procedure:

1. Identify the first quartile Q1 (the 25th percentile data point)
2. Identify the third quartile Q3 (the 75th percentile data point)
3. Identify the inter-quartile range(IQR): IQR=Q3-Q1
4. Calculate lower limits (L) and upper limits (U) by the following formulas:
   o L=Q1-1.5IQR
   o U=Q3+1.5IQR
5. Detect outliers by the rule: An observation is an outlier if it lies below the lower bound or beyond the upper bound (i.e. less than L or greater than U)

Data Structure
The completed dataset for the “Household registration survey 2015” includes 9 files in STATA format (.dta):
• hrs_maindata: Information on the households, including: assets, housing, income, expenditures, social inclusion and social protection issues, household registration procedures
• hrs_muc1: Basic information on the household members
• hrs_muc2: Education of the household members
• hrs_muc3: Healthcare status of the household members
• hrs_muc4: Employment situation of the household members
• hrs_muc7cc2: Pension and unemployment benefits or severance payment situation of the household members
• hrs_muc7cc4: Regular social allowance/benefit of household members
• hrs_muc9: Households’ financial obligation
• hrs_muc9c19: Social network of households

Data format in Stata has the following structure:
• Each column represents one variable. The variables in the dataset are named by the location of the question within the questionnaire (for example, m1c2 in the data file captures the answer to question 2, section 1 in the questionnaire). There are three types of variables: (i) discrete choice; (ii) continuous and (iii) text (or string) variable. Discrete variables have value labels and variable labels, while continuous and text variables have variable labels only.
• Each row represents one observation. For instance, each row captures information on a household, a household member or a loan.