C. Statistical Analysis
To look at program effects into the IPV and you will paths, the analysis rooked the usage a beneficial cutoff tolerance put on a continuous system qualifications index (PMT rating) to identify cures and you can evaluation communities. Hence, results are appropriate with the populace in the tolerance and you may create not depict an entire distribution out-of socioeconomic position in the analysis teams. Since an excellent discontinuity is used on the task of procedures, i make certain brand new conditions that generally need keep for a routine regression discontinuity framework approach to feel appropriate within analysis, in addition to no control within threshold otherwise expertise in the new formula from the possible beneficiaries to influence its qualification updates. Brand new certified take to of balance anywhere between analysis and you will medication teams are including stated to the latest take to from inside the dining tables 2 and you can 3 (col. 10). Regarding total centering on exercise used, brand new PMT scores varied away from ?step three.294 to cuatro.601 z-ratings, throughout the research shot, the newest PMT ratings cover anything from ?0.571 so you can 0.324 z-scores, on the system eligibility cutoff on ?0.106 z-ratings ( practical departure = 0.32 ). 6 Home were systematically sampled in the cutoff to own evaluation purposes, while the resulting domiciles on treatment try is actually within 0.46 practical deviations of your own cutoff, and all evaluation property is actually inside 0.43 basic deviations of the cutoff. Given the small endurance of your own shipment out of tested houses doing the latest PMT score, a straightforward Did was applied to test the fresh new impacts of your program: eight
Manage variables used in the study integrated individual- and household-peak signs and you may area-top repaired effects. 8 At the private level, i managed to have years (in many years) and you can indicators the certified training out-of one another girl and you will spouse. nine Further, in the research of your own full test, we manage to be in the a great polygamous relationship (site class was nonpolygamous otherwise, getting simple malfunction, a great monogamous commitment). Household-height handle parameters provided house proportions and PMT score. 10 I declaration the brand new a style of the region indication parameters inside detailed balance and you can attrition tables; yet not, while we were neighborhood-height fixed outcomes, Heterosexual dating dating service these symptoms are not used in regression data (districts: Eastern Mamprusi, Karaga, and you can Yendi on the Northern Area and you will Bongo and you may Garu Tempane throughout the Higher Eastern Area). To assess distinctions of the members of the family structure, we work together the fresh new signal to own polygamous union towards treatment indicator, big date sign, and feeling estimate (starting a triple communication title): 11
D. Attrition
Attrition over the panel period has potential to threaten the internal validity and generalizability of the findings. Overall attrition in our analytic sample is approximately 10%, slightly higher than the overall household sample rate of 6.6%, and does not vary significantly between treatment and comparison samples ( p -value = .35 using a simple t-test; Table 1). Further, we investigate whether the sample lost to follow-up differs significantly from our panel sample in terms of IPV outcomes, control variables, or pathway indicators (tables 2, 3). Column 7 shows the mean difference between the sample lost to follow-up in the treatment group, as compared with the same in the comparison group. Column 8 shows the p-value of this difference derived from a regression using treatment to predict the specific indicator, controlling for the PMT score and restricting to attritors. Table 2 shows that across 12 background characteristics and 14 pathway variables, only one is significant at the p < .05 level (there is a statistically significantly higher proportion of the treatment group lost to follow-up as compared with the comparison group in Bongo district). Table 3 shows that across 15 outcome indicators, the only ones to show evidence of differential attrition at p < .05 or lower are those measuring the frequency of sexual IPV (means of those lost to follow-up in the comparison group are slightly higher at .28 vs. the treatment group at .26). These results indicate that the potential for bias in overall findings due to attrition is low. Despite these promising results, we do find that women leaving our sample across both treatment arms differ in terms of background characteristics from those who remain in our sample (p-values shown in cols. 3 and 6). For example, women who leave the sample are generally younger, better educated, in monogamous partnerships (vs. polygamous), and live in smaller households. This may be reflective of rural-to-urban migration or other mobility patterns, although these differences are not generally observed for pathway variables, including indicators of economic standing.