7.6 Do ASHA - Beneficiary Differences Impact Service?

We examine if differences between ASHAs and Mothers limit the frequency or quality of their interactions. We do this with two statitistical approaches. In the first, we fit two regression models to make an initial estimate of what potential barriers to services there might be. One model looks at factors that correlate with the maternal health score and with the ASHA interaction score, separately.

In the following plot (7.11), we compare two models that differ in the response variables used. One is a maternal health score and the other is an ASHA interaction score. These are the same two variables depicted above in Figure (7.10) where maternal health score is shown to increase with ASHA interaction. As such, we examine if wealth, caste, or religion could impede either side of this relationship, the access to ASHAs or the uptake of healthy behaviors.

Caste and religious difference were calculated a similar way. Both are simply binary variables where a 1 means ASHA and her beneficiary are from the same caste or religion and a 0 means that they are from a different caste or religion. The wealth difference is a continuous variable that subtracts the wealth value of the ASHA from the wealth value of the mother. This is the same wealth measure described in Section 6.3.0.6.

This Figure (Figure 7.11) also shows what demographic factors influence each of these scores. Higher parity had a negative effect on health score. Education was positive and increasing for both variables, as was age. The main result is that ASHA interaction score may be higher when mothers and beneficiaries are from the same caste.

Assessing variation in ASHA interaction and uptake of health behaviors among Mothers, as a function of parity, fertility, education, and three measures of ASHA-Beneficiary difference: wealth, caste, and religion. Plot shows Incident Rate Ratios (similar to Odds Ratios) from a negative binomial generalized linear model. Health score is the sum of recommended behaviors adopted and ASHA interaction score is a sum of all ASHA interactions and mentions. Each model has the same predictor variables and differ only in the response variable.

Figure 7.11: Assessing variation in ASHA interaction and uptake of health behaviors among Mothers, as a function of parity, fertility, education, and three measures of ASHA-Beneficiary difference: wealth, caste, and religion. Plot shows Incident Rate Ratios (similar to Odds Ratios) from a negative binomial generalized linear model. Health score is the sum of recommended behaviors adopted and ASHA interaction score is a sum of all ASHA interactions and mentions. Each model has the same predictor variables and differ only in the response variable.

The model behind Figure (7.11) has some limitations. One issue is that the behaviors that are tallied in the Maternal Health Score have many influences and this model considers only differences between the ASHA and Mothers in her catchment area.

Further, here we are interested in the two variables plotted against each other in the previous section, the ASHA Interaction Score and the Maternal Health Score (Figure 7.11). Having an initial impression of how ASHA-beneficiary differences affect both the ASHA interactions and the uptake of health behaviors is useful, because it shows that the differences might affect both the input and the result of concern, but we might want to consider how these measures of difference each interact with ASHA interaction.

The second statistical analysis of the impact of ASHA - beneficiary differences on the amount of interaction is a moderation analysis. The following moderation analysis estimates how much (or how little) ASHA interaction is amplified or muted by each of the differences.

This table shows the results of a moderation analysis, whereby we consider if each different (wealth, religion, caste) independently changes the effect of ASHA interaction score on maternal health score. In conducting the analysis we found that the model had ‘under-dispersion,’ which means that conventional approaches to modeling count data might yield biased estimates. For this reason we fit a Conway-Maxwell Poisson generalized linear model. While that seems like a ‘mouthfull,’ all it does is account for cases when count data have less variance than assumed by the Poisson. This is less common than cases where there is more variance than assumed (over-dispersion).

Table 7.12: Moderation analysis for wealth, caste, and religion as factors possibily affecting the relationship between ASHA interaction score and maternal health score.
Wealth Caste Religion
(Intercept) 1.891*** 1.888*** 1.927***
[1.846, 1.937] [1.840, 1.936] [1.862, 1.991]
Parity2 -0.075** -0.073** -0.077**
[-0.124, -0.025] [-0.123, -0.023] [-0.127, -0.027]
Parity3 -0.120*** -0.124*** -0.127***
[-0.180, -0.061] [-0.183, -0.065] [-0.187, -0.068]
Parity4 -0.108** -0.114** -0.117**
[-0.182, -0.034] [-0.187, -0.041] [-0.190, -0.044]
Parity5+ -0.231*** -0.240*** -0.241***
[-0.323, -0.140] [-0.331, -0.148] [-0.332, -0.149]
EDU1to7 0.029 0.030 0.036
[-0.027, 0.085] [-0.026, 0.086] [-0.020, 0.091]
EDU8to10 0.085*** 0.095*** 0.096***
[0.040, 0.131] [0.050, 0.139] [0.052, 0.140]
EDU11to13 0.145*** 0.154*** 0.161***
[0.079, 0.211] [0.091, 0.217] [0.098, 0.224]
EDU14to17 0.213*** 0.220*** 0.227***
[0.141, 0.286] [0.151, 0.290] [0.158, 0.297]
Age21-24 0.018 0.020 0.023
[-0.032, 0.067] [-0.029, 0.070] [-0.027, 0.072]
Age25-28 0.055+ 0.064* 0.064*
[-0.008, 0.119] [0.002, 0.127] [0.001, 0.126]
Age29-33 -0.007 0.002 0.002
[-0.091, 0.077] [-0.082, 0.086] [-0.082, 0.086]
Age34+ 0.052 0.063 0.060
[-0.058, 0.162] [-0.047, 0.173] [-0.050, 0.170]
ASHA_Int_c 0.021*** 0.026*** 0.014***
[0.018, 0.023] [0.021, 0.031] [0.006, 0.022]
wealth_diff_c -0.011
[-0.025, 0.004]
ASHA_Int_c × wealth_diff_c 0.002
[-0.001, 0.004]
caste_diff1 -0.002
[-0.037, 0.033]
ASHA_Int_c × caste_diff1 -0.008**
[-0.014, -0.002]
relig_diff1 -0.047+
[-0.101, 0.006]
ASHA_Int_c × relig_diff1 0.007
[-0.002, 0.016]
Num.Obs. 1158 1158 1158
AIC 4830.6 4827.8 4830.0
BIC 4916.5 4913.7 4915.9
Log.Lik. -2398.304 -2396.886 -2397.978
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
Results of moderation analysis

Figure 7.12: Results of moderation analysis

The moderation analysis suggests that differences of both caste and religion moderate the relationship between ASHA interaction and maternal health score (Figure ??).

Differences in caste and religion as potential moderating effects.

Figure 7.13: Differences in caste and religion as potential moderating effects.

##                  OR 2.5 % 97.5 % Estimate Std. Error z value Pr(>|z|)
## (Intercept)   0.295 0.165  0.515   -1.220      0.291  -4.199    0.000
## Parity2       1.239 0.809  1.902    0.214      0.218   0.983    0.326
## Parity3       1.148 0.692  1.909    0.138      0.258   0.535    0.593
## Parity4       0.951 0.508  1.769   -0.050      0.318  -0.158    0.875
## Parity5+      1.128 0.537  2.334    0.121      0.374   0.322    0.747
## EDU1to7       0.691 0.416  1.108   -0.370      0.249  -1.487    0.137
## EDU8to10      1.104 0.759  1.599    0.099      0.190   0.524    0.600
## EDU11to13     0.885 0.476  1.581   -0.123      0.305  -0.403    0.687
## EDU14to17     0.654 0.296  1.324   -0.424      0.378  -1.121    0.262
## Age21-24      0.857 0.564  1.303   -0.154      0.213  -0.722    0.470
## Age25-28      1.108 0.663  1.853    0.102      0.262   0.390    0.696
## Age29-33      0.782 0.387  1.543   -0.245      0.352  -0.698    0.485
## Age34+        0.628 0.233  1.552   -0.465      0.480  -0.968    0.333
## wealth_diff_c 1.117 0.989  1.264    0.111      0.063   1.769    0.077
## caste_diff1   0.706 0.529  0.942   -0.348      0.147  -2.369    0.018
## relig_diff1   1.161 0.743  1.874    0.150      0.235   0.636    0.525