Chad Survey Methodology


Fieldwork

Fieldwork took place between September 13th and November 2nd, 2009, spanning approximately 50 days total.

Dates and Locations of Interviewing

Table 1 contains the exact dates and locations of interviewing, complete by each region and department or district

Table 1



Description of Sampling Design And Procedure


This section contains details of the universe sampled, including exclusions, the sample size, sample preparation, and the selection procedure from sampling points down to individual respondents.

Universe Sampled

The sample universe includes all citizens who meet the age requirement (15 or older) at the day of the interview.

The population source for Chad is the last census that was held in the country in 1993.  The population of 8,811,483 (excluding the Boukou Enneddi Tibesti region—see below; with this included it would be 8,915,381) is a projection for 2009 based on the official census data.  Whilst we do not have exact figures for the 15+ population in 2009, as of 1993, this group represented 52.1% of the population.

Due to inaccessibility (extreme distance and harsh climates), and thus exorbitantly high transportation costs, the Boukou Enneddi Tibesti region was excluded from the sample. This region represents 1.1% of the national population, and therefore excluding it from the sample would not have a significant impact on the representativeness of the survey.

Sample Size


Our final dataset includes 1,985 interviews.

Sample Preparation

We used a three stage stratified cluster design and distributed the base sample  among 13 out of 14 regions using PPPS sampling (proportionate to population probability sampling). As such, larger (i.e. more populated) geographical units had a proportionally greater probability of being chosen.  Please note that although Chad currently has 18 regions, the most recent census data is prior to the addition of 4 regions; our sample is drawn off of the 14 region census, of which we neglected to sample one region due to time, logistic and security issues.  Please see Table 2 for further details concerning the sample allocation.



Selection Procedure

The following selection procedure was used to identify starting points and select households and respondents:

  • To identify the starting points, the supervisors prepared a sketch map of the area, identifying main boundaries, streets, and landmarks.  This was then divided into non-overlapping sectors; subsequently, specific sectors in which to carry out the survey were randomly selected.
  • The blocks within these sectors were listed and then also selected randomly as a starting point for the interviewer’s route. 
  • From the starting point, the first house was selected using the day code, with a skip pattern of every 5th house in an urban area and every 3rd house in a rural area implemented to choose each subsequent house.
  •  Once a household was selected, the kish grid was used to identify the respondent.

Starting Points

In each location, suitable landmarks or streets were selected as starting points from which the interviewers would depart in different directions to begin their household selection. 

Weighting Procedures

A 2-stage sequential post-stratification weight was applied to the data.

First, the sample was weighted according to the urban/rural distribution of each region.  Weights were needed in each region to bring it back to its true national representation of urban/rural divides. From the sampling frame, a cross-tab of region by habitat (urban/rural) for the universe (N) was used.  A similar cross-tab for the sample (n) using the database was also used to be compared with the Universe and hence calculate the weighting factors

After applying the urban/rural weight, we ran the frequency distribution according to gender.  Due to some cultural difficulties  in interviewing women, our sample was approximately two-thirds men and one-third women, which differs from the national proportion of 48% men and 52% women.  From this we were able to calculate the appropriate gender weight.

We then multiplied the two weights together and applied the weight to each region separately for urban and rural men and women.

The separately calculated weights are located in Tables 4 and 5.



Sample Error
The maximum error for this sample (1985) at a confidence level of 95% is +2.2%.