Does Insurance penetration affect care seeking for Under-5s with diarrhoea? evidence from the Multiple Indicator cluster Survey in Nigeria

Ikpeme Neto
3 min readNov 9, 2024

--

I’ve recently began learning econometric techniques to help me understand and better analyse data in order to provide evidence for policy interventions in Nigeria. To keep track of my thoughts and learnings, I’ve decided to keep working blogs for the analyses I run. These blogs are well below the standard required for proper scientific enquiry and publications and may well contain errors, ( you’re welcome to point them out, I’m still learning), but they will help me evolve my thinking, hypothesis and skills around the topics I’m interested in. Hopefully it might also help inform and inspire other interested folks.

Today, the simple idea I’m interested in is whether there’s any basic evidence that the more insurance adoption there is in a state, the more likely it is for children under 5 with symptoms of diarrhoea to seek for care. The data is taken from Nigeria’s multiple indicator cluster survey 2021. This survey provides penetration rates for health insurance by state for children under-5.

and rates of diarrhoeal symptoms in the preceding 2 weeks for the same under 5 age group, by state

The null hypothesis (H0) is that Insurance adoption does not significantly influence care-seeking behaviour of parents for their children.

A simple linear regression model for this is shown below

CareSeeking = β0 ​+ β1​⋅Insurancepenetration + u

While there are clearly other factors that likely affect this relationship, I’ve decided to keep this regression simple to see if there’s any statistically significant relationship in the first place before attempting to control for other factors.

I extracted and formatted the data in excel then imported it into Stata. Six states didn’t have reported figures for some of the variables so I omitted them from the analysis. To make it easier for me to understand the regression result, I generated a new variable of “percentage of children where advice or treatment was sort”. I did this by simply subtracting the percentage of children where “no advice or treatment was sought”, the variable given in the survey, from 100. See command below

gen tagesoughtcare2 = 100 - noadviceortreatmentsought

summary statistics for the data set is below:

sum

Variable | Obs Mean Std. dev. Min Max
-------------+---------------------------------------------------------
state | 0
noadviceor~t | 29 28.64138 12.80565 2.4 57.3
noofchildr~n | 29 111.4138 114.0196 18 475
tagecovere~e | 29 2.765517 4.345875 .1 19.9
numberofch~c | 29 19.44828 27.60381 1 124
-------------+---------------------------------------------------------
numberofch~5 | 29 844.5517 476.8542 274 2434
tagesought~2 | 29 71.35862 12.80565 42.7 97.6

The mean percentage of care seeking for children with diarrhoea was 71% with a min and a max of 42% and 97% respectively.

The mean percentage of children with insurance was 2.7% with a min and a max of 0.1% and 19.9% respectively

I then ran the regression:

. regress tagesoughtcare2 tagecoveredbyanyinsurance

Source | SS df MS Number of obs = 29
-------------+---------------------------------- F(1, 27) = 1.59
Model | 255.665213 1 255.665213 Prob > F = 0.2178
Residual | 4335.90515 27 160.58908 R-squared = 0.0557
-------------+---------------------------------- Adj R-squared = 0.0207
Total | 4591.57036 28 163.984656 Root MSE = 12.672

--------------------------------------------------------------------------------
tagesoughtca~2 | Coefficient Std. err. t P>|t| [95% conf. interval]
---------------+----------------------------------------------------------------
tagecoveredb~e | .6953118 .5510637 1.26 0.218 -.4353775 1.826001
_cons | 69.43572 2.803579 24.77 0.000 63.68325 75.18819
--------------------------------------------------------------------------------

The very low R-squared suggests a poor fit in the model. That is to say that there’s no relationship between care seeking and insurance adoption rates. Moreover, the t score for this regression is 1.26 with a p value of 0.218, suggesting the result is not statistically significant.

So going by this crude analysis, the null hypothesis is not disproved. The data from the MICS survey doesn’t provide evidence for care-seeking being affected by insurance adoption.

On to my next analysis.

--

--

Ikpeme Neto
Ikpeme Neto

Written by Ikpeme Neto

I build and write about companies, communities and culture

No responses yet