EquityTool: Released October 13, 2022

Source data: Djibouti Household Survey 2017

# of survey questions in full wealth index: 41

# of variables in full index: 105

# of survey questions in EquityTool: 11

# of variables in EquityTool: 16

Questions:

^*Office National de l’Eau et de l’Assainissement de Djibouti (ONEAD)

Technical notes:

Creating a full wealth index

To produce this EquityTool, we used the World Bank’s 2017 Djibouti Household Survey. While the survey’s final report contained indicator data stratified by wealth quintile, a wealth quintile variable was not included in the underlying data. Since a survey’s full wealth index is the comparator against which we evaluate the performance of an EquityTool, it was necessary to first create a full wealth index from the available household asset variables.

Simplification

We were unable to achieve sufficient agreement at the national and urban levels between the full wealth index and a simplified index using our standard simplification process (detailed in this article). Using a revised approach, detailed below, we achieved high agreement (kappa ≥ 0.75) for the national, urban, and rural indices.

The national factor weights used in the standard approach come from an analysis of the national population and contain only those variables that are related to the construct of wealth in the same way in both rural and urban areas. The national factor weights are usually used in EquityTools to calculate national quintiles, as they reduce some known areas of respondent error in the survey.

However, to overcome the problem of low agreement using the standard simplification approach, we instead used factor weights from the rural and urban analyses, which select variables that related to wealth differently in urban and rural areas. For example, in an urban area, ownership of goats may be more strongly associated with being poor than in rural areas. This is the case in Djibouti. A concise list of variables, common to both urban and rural areas, were iteratively selected to find those which result in high agreement (kappa ≥ 0.75) against the full wealth index quintiles for national, urban, and rural populations.

A score from the simplified index for urban residents (Uscore) was regressed against the wealth index score variable created for the corrected full wealth index analysis (Nscore), the same was done for rural residents (Rscore), and the resulting coefficients are used to create a single national score (NatScore).

Nscore=b1Uscore + a1

Nscore=b2Rscore + a2

NatScore=b1(Uscore)(Urban)+ a1(Urban)+b2(Rscore)(Rural)+a2(Rural)

Where Urban=1 if respondent lives in an urban area and 0 if otherwise, and Rural =1 if respondent lives in a rural area and 0 if otherwise.

Respondents’ quintile assignments resulting from NatScore, the national wealth index score created from a simplified list of questions were compared to the quintile assignments resulting from the full wealth index with 41 variables using the kappa statistic.

The questions in the simplified index which resulted from this process differ from EquityTools that are created using our standard approach. Notably, we need to know whether the respondent lives in an urban or rural area, thus an additional question has been added to the EquityTool for Djibouti: ‘Determine if the respondent lives in an urban or rural area’. In principle, the definition of ‘urban’ and ‘rural’ should match the definition used in the Djibouti household survey. Typically, this definition is defined by the country. In practice, the user needs to decide how to determine if each respondent lives in an urban or rural area. Three approaches are presented below, with some notes on each. Whichever method is chosen, it should be uniformly applied across all surveys conducted.

1. Ask the respondent directly – ‘is your home in an urban or rural area?’ This relies on the respondent’s understanding of ‘urban’ and ‘rural’.
2. Allow the data collector to determine whether the respondent lives in an urban or rural area, based on available guidance. This will work best if the interviews take place in or very near to people’s homes, and if the data collectors can be trained on the same rules to determine if an area is urban or rural. One example of a rule is to classify ‘peri-urban’ areas on the edges of a city or town as urban. Another rule might be to classify an area as urban if it has a market center which operates daily.
3. If the interviews are taking place outside the home, then classify respondents based upon the location of the interview. For example, if interviews occur in health facilities, classify respondents as urban if the facilities are located in urban areas. Individuals may travel, so this method is also subject to error.

Level of agreement:

Respondents in the original dataset were divided into three groups for analysis – those in the 1^st and 2^nd quintiles (poorest 40%), those in the 3^rd quintile, and those in the 4^th and 5^th quintiles (richest 40%). After calculating their wealth using the simplified index, they were again divided into the same three groups for analysis against the original data in the full Djibouti Household Survey. Agreement between the original data and our simplified index is presented above.

What does this mean?

When shortening and simplifying the index to make it easier for programs to use to assess equity, it no longer matches the original index with 100% accuracy. At an aggregate level, this error is minimal, and this methodology was deemed acceptable for programmatic use by an expert panel. However, for any given individual, especially those already at a boundary between two quintiles, the quintile the EquityTool assigns them to may differ to their quintile according to the original Djibouti Household Survey wealth index.

The graph below illustrates the difference between the EquityTool generated index and the full Djibouti Household Survey wealth index. Among all of those people (20% of the population) originally identified as being in the poorest quintile, approximately 94% are still identified as being in the poorest quintile when we use the simplified index.  However, approximately 6% of people are now classified as being in Quintile 2.  From a practical standpoint, all of these people are relatively poor. Yet, it is worthwhile to understand that the simplified index of 11 questions produces results that are not identical to using all 41 questions in the original survey.

The following table provides the same information on the movement between national quintiles when using the EquityTool versus the original Djibouti Household Survey wealth index:

The following graph provides information on the movement between urban quintiles when using the EquityTool versus the original Djibouti Household Survey wealth index:

The following table provides the same information on the movement between urban quintiles when using the EquityTool versus the original DHS wealth index:

Data interpretation considerations:

1. This tool provides information on relative wealth – ‘ranking’ respondents within the national or urban population. The most recent available data from the World Bank indicates that 17% of people in Djibouti live below $1.90/day [1]. This information can be used to put relative wealth into context.
2. People who live in urban areas are more likely to be wealthy. In Djibouti, 23.4% of people living in urban areas are in the richest national quintile, compared to only 1.1% of those living in rural areas [2].
3. If your population of interest is predominantly urban, we recommend you look at the urban results to understand how relatively wealthy or poor they are, in comparison to other urban dwellers.
4. If the people you interviewed using the EquityTool live in rural areas, or a mix of urban and rural areas, we recommend using the national results to understand how relatively wealthy or poor they are, in comparison to the whole country.
5. Some regions in Djibouti are wealthier than others. It is important to understand the country context when interpreting your results.
6. In most cases, your population of interest is not expected to be equally distributed across the five wealth quintiles. For example, if your survey interviewed people exiting a shopping mall, you would probably expect most of them to be relatively wealthy.

Metrics for Management provides technical assistance services to those using the EquityTool or wanting to collect data on the wealth of their program beneficiaries. Please contact support@equitytool.org and we will assist you.

[1] From povertydata.worldbank.org, reporting Poverty headcount ratio at $1.90/day at 2011 international prices.

[2] From the Djibouti Household Survey 2017 dataset, available with permission at https://microdata.worldbank.org/index.php/catalog/3463/study-description