Page  2

Targeting

Almost all program planning involves identifying who the "worse-off" people are, and knowing where they are located.  Ranking populations by prevalence of malnutrition, mortality, access to services, or environmental factors, is a device used to determine the "worst-off" population. Mapping often helps to display such results. Ranking involves straightforward one-way analysis, showing the outcome indicator by one classifying variable at a time: prevalence of low arm circumference by district, for example.

Breaking down the national situation analysis (like Tanzania, shown earlier) by region, district, etc., gives exactly the sort of information required, with the proviso that some further selection of outcome indicators may be needed.  At an area level you can’t easily use 10 or 20 indicators, as is typically and usefully presented at the national level.

Targeting often uses ranking analysis, where it provides a comparison between geographic areas (or other grouping) for an outcome variable ranked by prevalence, by access to services, or some other factor of interest. This can help in program planning to assign the interventions to the areas with the worst outcome and the least current access to assistance.


RANKING ANALYSIS

With SPSS it is possible to sort your outcome variable, such as prevalence of low arm circumference, in descending order and then print out Case Summaries with several indicators. This will give you a quick look at possible relationships and give you ideas for further analysis.

Take a look  at how to create multi-indicator descriptive analysis or RANKING ANALYSIS for Bangladesh data.

wpe3.jpg (25908 bytes)

INTERPRETATION:

The output was limited to the first 10 districts after sorting by arm circumference (highest percentage of low arm circumference). The relationships between the malnutrition level (estimated by low arm circumference) and determining variables such as education, health care access (% measles immunization), and socio-economic status (roofing type) are not very clear.  The district with the highest malnutrition, Lakshmipur, also has a relatively high percentage of health access, education and quality roofing.   The associations can be visually estimated as you scroll your eyes down the list.   It is clear that there are inconsistencies in the results from what would be expected, but this does not mean that these factors are not influencing the nutritional status of the child. The associations between influencing factors and outcome variables may not be perfectly discernable using the ranking analysis (especially at the aggregated level), but ranking can still be useful for targeting the areas that are worse off. 


TARGETING PRIORITIES: Should We Target By Number of People Affected or By Prevalence?

Often  programs are established and maintained with limited resources.  In these situations there should be proper targeting to areas with the most need.  This raises the issue of targeting by number of people affected versus by prevalence. The goal of  interventions is to affect the largest number of people. If a limited number of intervention sites are established, then there needs to be appropriate consideration of which areas have higher percent of illness in the population and which areas have more cases, simply because the population of the area is large. Keep in mind, the goal is to reach the most in need.

HINTS:

Usually rely on % or the prevalence, but take some account of large populations with medium prevalence.
Consider whether expenditures are fixed by area, or per head, or a balance between these.
Most program expenditures are more per head than fixed by area; hence, target by prevalence.

EXAMPLE OF TARGETING BY NUMBER OR %

District 1: Child population =100,000
Prevalence malnutrition = 50%
Number malnourished (Pop x Prev) = 50,000
District 2: District 2: Child population =500,000
Prevalence malnutrition = 20%
Number malnourished (Pop x Prev) = 100,000

If costs are per head: say $2 per head
District 1 costs $200,000; hence $4 to reach each malnourished child
District 2 costs $1,000,000; which is $10 to reach each malnourished child
In this situation, the area with the highest prevalence is the best area to target. If it is possible to perfectly target where the cost is the same despite transport of goods, location, etc... then it would be possible to just pick up the malnourished children along the way. This is not usually the situation, unfortunately.

If costs are fixed by area: say $1 million per area
District 1 is $20 to reach each malnourished child.
District 2 is $10 per malnourished child.
In this case, District 2 is the better choice because more children can be reached with this scenario, despite the higher prevalence in District 1. This scenario rarely applies though. Most often targeting goes to the highest prevalence areas.

If a mix of fixed and per head costs: it depends on the ratio of the two.

 


The Steps to Targeting:

1.  Rank districts by Prevalence

2.  Add number and % of malnourished

3.  Rank by number of malnourished

4.  Note which districts go up to near the top of the ranking

5.  Rank again by prevalence and make initial selection by prevalence, but taking some account of priority districts from step 4-   JUDGEMENT IS NEEDED!

6.  Add indicators of health service

 

Take a look at how to use SPSS in the  STEPS TO TARGETING  with district prevalence of low arm circumference in bdeshc.sav.

The resulting output at step 6 is shown here. The output reveals that there are differences in which districts have the highest prevalence of malnutrition, the greatest number of malnourished children, and the least access to health care.  As discussed previously, it is important to look at all of these factors and to consider how to best fulfill the demands of a funding agency and reach the most individuals in need.  There is not a simple answer and each situation will need to be treated with special care, but the goal is to most effectively and efficiently reach the most children in need. 

wpe1.jpg (35108 bytes)

Try an  EXERCISE IN TARGETING using the bdeshc.sav data set from Bangladesh.


WEIGHTED MEANS: Prevalence of Malnutrition

When the number of children vary between districts and you want to know the prevalence of malnutrition overall, it is important to be able to calculate the weighted average of district data.  This can be done in two ways, depending on the information available. 

Given the following district level information calculate the prevalence of underweight for the province.

Child Population

Prevalence Underweight (%)

# Underweight Children*

District A

10,000

40

?

District B

40,000

20

?

Province

50,000

?

?

        Step 1: Calculate the number of underweight children using the child population and underweight prevalence
                    - i.e. multiply 10,000 by 0.4 to get 4,000 underweight children. 

        Step 2: Then calculate the prevalence of underweight using the total number of underweight children and the total child population
                    - i.e.divide 12,000 by 50,000 and then multiply by 100% to get a regional prevalence of underweight of 24%. 

Child Population

Prevalence Underweight (%)

# Underweight Children*

District A

10,000

40

4,000

District B

40,000

20

8,000

Province

50,000

24

12,000

When the only information given is the total population and underweight prevalence data for each district then the process is slightly different. 

Given the following district level information calculate the prevalence of underweight for the province.

 

Child population as a % of provincial child population

Prevalence Underweight (%)

District A

60

40

District B

40

20

Province  

?

        Step 1: Given only child population as a percent of the total population and the underweight prevalence, use both valued as a percentage.

       Step 2: Multiply each percent prevalence underweight by its weighting factor (percentage of child population) and add the two together. (0.60)(0.40) + (0.40)(0.20) = 32

 

Child population as a % of provincial child population

Prevalence Underweight

District A

60

40

District B

40

20

Province  

32


   CURRENT PROGRAM TARGETING: Service Coverage of Nutrition Related Programs

When making recommendations for nutrition related programs it is important to assess targeting of current programs, and perhaps retarget current resources as opposed to creating a new program.  In the Eastern Kenya data set, if we run a compare means with prevalence of underweight as the dependent variable and recorded measles immunization as the independent we get the following output:

 

coverage.jpg (21526 bytes)cov2.jpg (14554 bytes)

                                               

From this we can then create a dummy table in the following manner:

  Yes <-2SD's Not <-2SD's Total
Yes, measles shot

?

?

503
No, measles shot

?

?

177
Total     680

From the above output we can fill in the empty spaces of our dummy table.  For example, 22.03% of those children not in a vaccination program are malnourished (.2203 X 177 = 39) while 32.80% of the children participating the vaccination program are malnourished (.3280 X 503 = 165).  Simple subtraction allows us to complete the table.

  Yes <-2SD's Not <-2SD's Total
Yes, measles shot

165

 

503
No, measles shot

39

 

177
Total 204   680

Simple subtraction allows us to complete the table (503 - 165 = 338  and 177 - 39 = 138):

Yes <-2SD's Not <-2SD's Total
Yes, measles shot

165

338

503
No, measles shot

39

138

177
Total 204 476 680

This 2 X2 tables helps us figure out the following:

Population Prevalence: prevalence of underweight in the population {(a+c) / (a+b+c+d)} 
204/680 = 30.0%%
Population Coverage: the proportion of the whole population participating in the program {(a+b) / (a+b+c+d)}
503/680 = 73.9%
Malnourished Coverage:  the proportion of the malnourished children participating in the program  {a / (a+c)}
165/204 = 7.4%
Focusing (Program Coverage):  proportion of the those in participating in the program who are malnourished  {a/ (a+b) }
165/503 = 32.8%
F/PP Ratio:  A way to test for targeting is to determine the ratio of those in the program who are malnourished (focus)
to the population prevalence (PP).  [a/(a+b)] / [(a+c)/(a+b+c+d)]  If the ratio is greater than one, then program
targeting is happening.
       F/PP = (165/503) / (204/680) = 1.09

INTERPRETATION

Thirty percent of the population is underweight and only 7.4% of those participate in the program.  The malnourished, however, make up 32.8% of the children who are participating in the program. By looking at these results, we can see that the   measles vaccination program is well targeting towards those worse off.


RE-TARGETING: Techniques and Program / Service Coverage

1.  In order to assess targeting and program delivery, it is useful to examine outcomes such as prevalence of low arm circumference or diarrhea in relation to indicators of relevant interventions such as access to service. For example, in the Bangladesh data set, we can look at the scatterplot of diarrhea prevalence and prevalence of diarrhea treatment by a trained health worker.

Try an exercise to produce a SCATTERPLOT of the prevalence of diarrhea by treatment of diarrhea.

The basic relationship looks like this.

wpe2.jpg (15771 bytes)

INTERPRETATION:

This graph shows that there is not a strong relationship between diarrhea prevalence and access to trained health workers, but that is not what we are looking for.  The research question is:  "Which districts have a high prevalence of diarrhea and poor access to trained health workers?" and then:  "What are the prospects of shifting resources to ensure the high diarrhea prevalence districts get more equitable access to trained health workers?"  The districts with high diarrhea prevalence and low trained health worker coverage are cases such as 25, 11, 46, 2, 1, etc., found in the top left hand corner of the graph.

 

2.  The next step will be to define some cutpoints for high/low coverage and high/low prevalence, resulting in four categories. It is most useful to define these cut-points using the mean for each variable.

Try this exercise on DEFINING CUT-POINTS to create categories of illness and categories of health service.

wpe8.jpg (11199 bytes)

 

It is useful to roundup from the means to logical cut-off points. In this case we will roundup and use below 15% diarrhea prevalence and below 20% health worker coverage as our cut-offs.

 

3. Dichotomizing the variables using these cut-offs will make it possible to separate those that are most in need and also make it easier to present the situation in a two-by-two table.

Try an exercise on DICHOTOMIZING THE VARIABLES.

wpe9.jpg (7444 bytes)

In this table, there are two categories for the diarrhea treatment by a health worker (HLDITRHW:  represents a high/ low diarrhea treatment by health worker) that are coded 1 for <20% (low) and 2 for >=20% (high).  There are also two categories for prevalence of diarrhea (HLDIAPRV:  represents a variable high/ low prevalence of diarrhea) that are coded 1 for <15% (low) and 2 for >=15% (high). It is easy now to see where the deficiencies are in targeting, and to re-target the healthworkers from the high coverage/ low prevalence areas to those areas that are high prevalence/ low coverage. This will be the covered in the next lesson.


PRESENTING THE INFORMATION

Presenting this information with Microsoft Word or other word processing package might result in more attractive 2x2 tables, if it is important to have presentable information. Also notice that the cells have been shifted so that the upper left cell is the high diarrhea prevalence and low health worker coverage districts (lower left above).

Diarrhea treatment by health worker * Diarrhea prevalence categories Crosstabulation

   

wpe8.jpg (18939 bytes)

  • According to this information, is this program well-targeted? 
    WARNING:  in order to "blindly" use the equations you must set up your 2X2 tables so that box A represents high prevalence of diarrhea and high coverage;  box B is low prevalence, high coverage; box C high prevalence, low coverage; and box D is low prevalence and low coverage.
  • You may want to re-do the presentation in the following manner to aid in the calculation of the program coverage.
          Diarrhea Prevalence
Health Worker   (Program)   High Low Total
High 5 (8%) 17 (27%) 22 (34%)
Low 14 (22%) 28 (44%) 42 (66%)
Total 19 (30%) 45 (70%) 64 (100%)
  • To determine if the program is well targeted, perform the following calcualtions:

    Population Prevalence:
    prevalence of underweight in the population {(a+c) / (a+b+c+d)} 
                                                   PopP= 29.69%
                                    
    Population Coverage: the proportion of the whole population participating in the program {(a+b) / (a+b+c+d)}                         PopC = 34.38%

    Malnourished Coverage:  the proportion of the malnourished children participating in the program  {a / (a+c)}                         MalC = 26.32%

    Focusing (Program Coverage):  proportion of the those in participating in the program who are malnourished  {a/ (a+b) }    ProC = 22.73%

    F/PP Ratio:  A way to test for targeting is to determine the ratio of those in the program who are malnourished (focus)          F/PP = .7655 (<1 is poor targeting)
    to the population prevalence (PP).  [a/(a+b)] / [(a+c)/(a+b+c+d)]  If the ratio is greater than one, then program
    targeting is happening.
  • This information can then be used to re-target services from areas of low diarrhea prevalence and high health worker coverage to areas of high diarrhea prevalence and low health worker coverage.

Retarget Health Workers

wpe6.jpg (23478 bytes)

The health workers from the blue cell are in an area where the need is low, but the coverage is high.  These health workers would be better used in an area where more children are needing care (high prevalence), therefore the workers from 14 of the 17 districts with low prevalence/ high coverage move (indicated by the blue arrow) to the high prevalence/ low coverage areas (which would make them high prevalence /high coverage areas).  The 14 districts with low prevalence / high coverage that sent their health workers to high prevalence / low coverage areas are now reclassified to low prevalence and low coverage areas (since they have sent their health workers to help others).  So 14 districts are reclassified as low /low and that makes a total of 42 low/low districts (indicated by the white arrow on the right) and the 14 districts that receive health care assistance are now reclassified as high prevalence/ high coverage districts, which makes a total of 19 districts high/ high (indicated by the white arrow on the left).

 

By re-targeting health worker coverage to areas with high diarrhea and low coverage, these areas become high diarrhea/high coverage. It is optimal to have 0 districts with high diarrhea and low coverage. In this case, health workers were redistributed from 14 low prevalence/high coverage districts to 14 high prevalence/low coverage districts. The results are shown below:

After retargeting the health workers:

wpe7.jpg (20171 bytes)

RE-TARGETING HEALTH CARE WORKERS exercise will give more details on redistributing the health workers between high and
low coverage areas. 

Return to Top of Page