Page 78 - IJPS-11-4

International Journal of
Population Studies Droughts and intimate partner violence

We also incorporated the degree of autonomy enjoyed 2.5. Statistical analysis
by respondents, measured by the women’s autonomy We analyzed the associations between the variables using
index. DHSs ask whether women have a say in various the proportions test, Chi-square tests, Goodman and
decision-making processes. A response of “yes” is coded Kruskal’s gamma, Kendall’s tau-b test, Cramer’s V test,
as 1, and “no” is coded as 0. The autonomy index is the and kernel density plots. Furthermore, the DID technique,
sum of these responses, ranging from 0 to 6. The higher along with logistic regressions, was employed to estimate
the score, the greater the respondent’s autonomy in the effect of drought on IPV. All statistical analyses were
household decisions. Another control variable used is conducted using Stata 18.
the number of control issues faced by the respondent,
as reported by the DHS. This variable also ranges from 2.6. Estimated model
0 to 6, with a higher score indicating a more controlling We estimated the following model:
partner.
O =β +β Year+β drought +β (Year * drought )+β C +ϵ
We controlled for the partner’s current age (a categorical it 0 1 i 2 it 3 it it 4 it it
variable with eight categories), education level, and alcohol Where O represents the log odds of a woman facing
it
consumption. In addition, we controlled for household- IPV in time t, year is a binary variable, taking the value
it
level characteristics, including the respondent’s religion 0 of observation from NFHS-4 (2015 – 2016) and 1 for
(Hindu, Muslim, or Others), social group, household NFHS-5 (2019 – 2021), drought is a binary variable coded
it
size, urban residence, urban area, and household wealth as 1 for states experiencing the 2018 drought and 0 for
index. The household wealth index is a composite measure the rest of the states, and C represents a vector of control
it
of a household’s cumulative living standard, reflecting variables.
ownership of various consumer items, such as television, The main coefficient of interest is β , which captures the
3
housing type, toilet facilities, and drinking water sources. interaction between the year and drought. This coefficient
The wealth index categorizes households into five groups: reflects the drought’s impact on IPV, estimated using the
poorest, poorer, middle, richer, and richest. DID approach. ϵ represents the error term.
DHSs are cross-sectional in nature. Although we do 3. Results
not know whether the same households or clusters were
repeated across the two survey rounds, we know that We first present the descriptive statistics of the study sample
the sample is drawn from each district, meaning that the in Table 1. We then proceed to examine the similarities and
districts are repeated. We used this information to include differences in the incidence of IPV between the treatment
district-level fixed effects and cluster standard errors at the and comparison groups. In addition, we analyze how these
district level in all models. two groups have fared over time in terms of IPV incidence
(Table 2).
2.4. Study design
Next, we explored the change in the controlling behavior
The occurrence of NEM droughts in only some Indian of partners over time and analyzed the association between
states makes this study a natural experiment. Our study control issues faced by the respondents and the incidence
follows a “pre–post, with–without” design. The states of IPV. Figure 1A-C illustrate these findings. Finally, we
that were exposed to NEM drought were compared with estimate the effect of the NEM drought on IPV (Table 3
those that were not. This creates the “with” and “without” and Figure 2).
groups. The data are drawn from 2 time points: 2015 – We also conducted a subsample analysis by examining
2016 and 2019 – 2021. As there was no NEM drought in urban and rural samples (Table 4). The results align
2015 – 2016, this period is labeled “Pre” (pre-exposure to with our main findings. To verify the robustness of the
drought), whereas data for 2019 – 2021 represents “Post” estimates, we perform falsification tests by modifying the
(post-exposure to drought). drought timeline (Table S7). These results confirm that
We adopted natural experiment terminology to drought significantly affects IPV, and our main findings
describe our groups, referring to the 2015 – 2016 data are not mechanical.
(from NFHS-4) as “pre” and the 2019 – 2021 data (from
NFHS-5) as “post.” States not exposed to the NEM drought 3.1. Descriptive statistics
served as the “comparison group,” whereas those exposed Table 1 presents the descriptive statistics of the study
(Andhra Pradesh and Karnataka) were the “treatment sample. The total number of observations for 2015 – 2016 is
group.” 13,908 in the comparison group and 2215 in the treatment

Volume 11 Issue 4 (2025) 72 https://doi.org/10.36922/ijps.3065