Chapter 7

PANDA Final Project - Bringing everything together

Computing Routines to be used in testing:

Descriptive Statistics
Frequencies
Min/Max, Mean
Histograms
Also want to address histograms vs. bar charts and drawbacks of both
Scatter Plots-including curve fit
Compare Means
Correlation tables
Linear Regression

Combining all of the analysis skills you have learned and used in PANDA, devise a nutrition intervention in selected districts in the Western and Nyanza Province of Kenya. Analyze the DHS 1993 data (knywes2.sav) and prepare a report with recommendations and analytical justification, including tables and figures. Make sure to consider the following topics:

1)  Area profile (descriptive stats, SES and environmental characteristics)

First, create an area profile for the whole region. Include information on relevant variables, for example: prevalence of underweight for the region, % of mothers with low education, % of households without latrine access, % of households with a poor drinking water source, etc. Then stratify by district and create an area profile for each district, these numbers will be useful in the next question. This gives you an idea of the situation in the region and within districts, so you can see if there is one district that is particularly bad-off, or if there is one variable that is particularly low. It might be useful to group variables according to a broad category (i.e. SES proxy, environmental, health care access etc.). See chapters 1 and 2 if you have any questions.

2)  Targeting priorities for future programs: selected indicators by district and their implications.

Take the district information generated in #1 and rank districts according to prevalence of malnutrition, and then add in other variables like: total number of malnourished children, percent of malnourished, immunization coverage, mother's education/literacy. It will be useful to keep the districts ranked by prevalence of undernutrition, and then just add the other variables and check to see if the situation is consistently poor in districts with high prevalence and vice versa. Note that a district may not be considered to be bad-off based on its prevalence of underweight, but if the population is large, there may be a greater number of malnourished kids there (it may contribute a significant proportion of the total malnourished children). See chapter 3 for questions.

3)  SE and environmental factors-water and sanitation in the region (not by district), taking into account possible confounders such as other SE factors (education, housing and access to healthcare).

This is where the majority of your time will be spent. This question involves the stepwise analysis from one-way to multi-way to identify possible causal pathways. First, explore associations between single variables and the outcome variable of interest (WAZ & underweight prevalence). Make a table to summarize the data and include information on the size of the difference (between good and bad categories) and its significance. Based on findings here, then consider two-way associations. The best way to present the data here is to create a table for each association-a pseudo 2 x 2 table, then you can plot the values for ease of determining if there is interaction or not. The final step would be to perform linear regressions based on your findings in the first two steps, including interaction variables where they are significant, and controlling for confounding. Then, based on your analysis, you can conclude whether an independent variable is significantly associated with the outcome independent of other factors, and can make program recommendations according to this. Relevant chapters 3, 4 and 5

4)  Assessment of current programs-program coverage and targeting especially the            following variables, but also those related to possible future interventions.

health (immunization)

This question is meant to assess how well current programs are targeted to the people who are worse-off. Using data from the compare means routine, one can assess how effectively targeted programs are, whether there is a higher proportion of malnourished receiving the program or not. This can be determined by creating a dummy table, malnourished vs. receiving program, to summarize the number of people in each category. This information then can be used to compute the prevalence of underweight among the program recipients (f) and the population prevalence (pp). A well targeted program would have an f/pp ratio greater than one. You will want to explore variables in the above listed categories, as well as variables corresponding to program recommendations-perhaps retargeting will be sufficient, or at least a first step.

5)  Child feeding practices and how this would be addressed through your program.

Read through the sub-module on "analysis of child feeding patterns" in chapter 6 and create a chart in excel to summarize your results. Considering your results, you can usually only conclude that there is not enough exclusive breast-feeding, combined with early introduction of complementary foods. Therefore, some type of breast-feeding education component would be an appropriate intervention.

Note: Make sure to proceed stepwise through the analysis, it is a logical process and should be analyzed in this manner. The data set gives household level data, but data are not available for all children, therefore analysis will be done based only on child #1. Especially for #3, look at variables you are likely to be able to association in a 5 year program like water and sanitation or access to health care. Other variables like SES and education will be used more for controlling for confounding, but may be considered for longer interventions if warranted. Make programming recommendations based on your analysis and present your findings in tables, charts and text in the form of a report to present to an NGO or the Kenyan Ministry of Health.