Does Increasing the Share of Women in Parliament Lead to Higher Spending on Health in Public Budgets?

By Marie-Eve G. Augier

A generally accepted notion is that spending habits differ by sex: men lean towards spending more on luxury goods whilst women tend to spend more on health, education and the household’s basic needs. There is research to support this and many development projects have embraced the concept. For example, microfinance institutions like Grameen Bank in Bangladesh, only lends to women and conditional cash transfer programs like Prospera in Mexico, only makes payments to mothers. Thus, I wondered whether such a micro-level trend could translate to the macro-level and whether higher shares of women in parliament (WIP) could be associated with higher levels of public health spending (PHS) within a country.

That female politicians could strengthen government spending on human development is certainly a point that could support the fight for increased female political participation and the dismantling of structural barriers that prevent this. Increasing women’s political participation is important for several reasons. First is diversity, excluding women may prevent a country from choosing the best possible candidates. Second is the human rights and equality argument, all women have a right to equal standing with men, including the opportunity for political participation, and that right should be exercised fairly. Thirdly, the democracy argument: leadership of a true democracy should reflect the values of its entire people. As half the population, women’s voices must be included in national decision-making.

To investigate whether the share of WIP leads to increased levels of PHS, I consider the following:

(Public Health Spending) = α + β₁(Share of Women in Parliament) + β₂Z

In this equation, PHS as a share of GDP, the dependent variable, equals some constant α[1] plus β₁ times the share of WIP, the independent variable, plus β₂ times Z. The terms, β₁ and β₂, will indicate the relationship between PHS, WIP and Z, estimating their values is the ultimate goal of this exercise. In the above equation, Z represents the control variables selected for this model, which are expected to have some effect on PHS and WIP for which I need to account: GDP per capita, maternal mortality ratio and women’s rights.

GDP is a representation of healthcare system quality; more developed countries tend to have attitudes towards women that are more favorable, making political participation more acceptable. The maternal mortality rate, also a healthcare quality indicator, but also directly affects women’s ability to live full lives and participate in politics. The women’s rights variable reflects institutionalized attitudes towards women. Some of these will have a direct impact on PHS: right to contraceptive use, maternal health care etc. Because the post focuses on the effect of WIP on PHS, the other control variables have been grouped into Z. I took this data from the Quality of Governance dataset, covering a cross-section of 193 countries in the year 2010. Of those 193, 166 had values for all the data of interest.

Before attempting to estimate the relationship, I graphed PHS versus WIP to identify any potential relationship that may exist between the two highlighting outliers (See Figure 1). The dashed line is the lowess line, showing the local average of the values of PHS closest to a particular value of WIP. It gauges the relationship between two variables without requiring any assumptions be made about the data. The solid line is the simple linear prediction, and although they are different, the lowess line is close enough to it. Thus, I am confident in proceeding with the estimation as detailed in the formula above.[2]

Table 2 displays the results. The numbers are the estimates for the β’s associated with each variable. The estimate for β₁ in the final column is 0.034. Thus, for every 1-percentage point increase in WIP there is a 0.034 increase in PHS. This may seem small but consider a country with a trillion-dollar GDP. A 0.034 increase in its share of PHS would equal 34-billion dollars, a substantial sum. The asterisk next to it suggests that this number is statistically significant to the 0.05% level. This means there is only a 5% chance that this number would be estimated if zero were the true value of β₁.[3] Note that, across the different equations the estimated β₁ falls from 0.075 to 0.034, this occurred because as more factors are controlled for, less of the variation in PHS is attributed to WIP but it remains significant and large.

In conclusion, increasing the share of women in parliament could have a positive effect on public health expenditure. However, in all regressions there is the chance of reverse causality, i.e. the possibility that β₁ is significant because it is PHS causing WIP instead of the inverse as suggested by the hypothesis. However, consider Rwanda, which had a quota for female political participation imposed in 2003, and a country with a similar level of PHS and WIP prior to the implementation of the quota, Turkmenistan. After 2003, PHS surges in Rwanda, but remains low for Turkmenistan, further suggesting that it is a change in WIP that leads to change in PHS rather than the reverse (See Figure 2).

Appendix

Each variable was tested for skew, which is an uneven distribution around the average (see figure 3). The first graph with PHS versus GDP shows a lowess line with an inverted-U shape. This suggests that the term GDP² should be used. In the second graph showing PHS versus Women’s Rights, the graph is of a linear shape so no change is necessary. In the third graph, with PHS versus maternal mortality ratio, the graph has a positive skew, so instead of using the variable as is, a log transformation was performed. The fourth graph is PHS versus the new variable log of maternal mortality ratio. You can see that the lowess line of this graph is significantly straighter.

---

[1] α is also the value of PHS if WIP was equal to zero, i.e. it is the minimum value that PHS could be.

[2] The same exercise was performed for PHS versus each of the control variables. The results are discussed in the appendix, but based on those graphs I included the log of the maternal mortality rate and added a square term for GDP in the final conditional model.

[3] The estimated beta coefficients of the control variables were not explored in depth because as previously mentioned they were not the focal point of this exercise. However, it is important to note that with the exception of Women’s Rights, none of the other control variables were significant. Also important, is the jump in R-squared when moving from equation 1 to 3 in Table 2. This suggests that the last model is a much better fit than the first despite the fact that not all the coefficients are significant.