Effect of one independent variable of several dependent variables – best strategy?

Question

I have a question regarding which analysis strategy is best suited for our objective. In an exploratory study based on data from a survey we conducted ourselves in India, we are analyzing the consequences of a lack of suitable brides as a result of sex-selection. The determinants of sex-selection are well-researched and we ourselves also followed a separate objective where we analyzed these in separate analyses and regressions. What the consequences of a lack of suitable brides might be is still only limitedly researched as there is a lack of data.

This means that we are essentially looking at an independent variable and are interested in exploring what other variables/aspects it has an effect on. We find bivariate correlations that make a lot of sense theoretically even though they are generally not very strong. Our variable of interest is a binary variable of whether there is a shortage of suitable brides as a result of sex-selective abortions.

My main question is; is there any other suitable analysis strategy we could use where we are not simply limited to bivariate correlations?

Furthermore, for some of the correlations we find there are issues of directionality and endogeneity. One example is crime, where crime could be a reason for parents to prefer and sex-select sons as sons could be thought to offer protection in an insecure environment where daughters could instead be perceived as liabilities. However, crime could also be a consequence of a shortage of brides and we do find such a correlation. From our separate analyses of the determinants of sex-selective behavior where we used logistic regressions with odds of births being sons as outcome, we know that crime is not a determinant of sex-selection. Our interpretation of this is that crime is to a certain degree a consequence of bride shortages but not a “cause” of sex-selective abortions. However, one of our variables that is negatively correlated with bride shortage is a measure of restrictions of women’s mobility, i.e. a measure of disempowerment of women. This suggests that a shortage of brides is related to a better situation for women. The difficulty is that this same variable also has, unlike our other variables, a significant effect on the odds of births being sons (once again from other regressions). Its effect on the odds of births being sons is positive, however, whereas it is negatively correlated with our variable for bride shortages. Our interpretation of this is that greater disempowerment of women is related to greater likelihood of sex-selection but that less disempowerment (and thus a better ‘situation’ for women) is a consequence of a bride shortage (and thus by extension a consequence of sex-selection)?

My second question is; are we correct in drawing these conclusions? Is there a better way for us to make sure we are drawing the right conclusions?

I would be very grateful for any help we could get.

KML

Answer 1

Given that all your data is observational (necessarily so) and further given the complex nature of cause-effect in your data (which you show in your question) I would be very careful to avoid words like "cause" and "effect" in interpreting your results.

You state

Our variable of interest is a binary variable of whether there is a shortage of suitable brides as a result of sex-selective abortions.

I don't see why this is binary. Surely the shortage of females can be mild, moderate, extreme ... and can be indexed by the male-female ratio in a given age range?

Then this variable (male-female ratio at suitable age range) could be used as a dependent or independent variable in regressions, which would allow you to control for multiple independent variables at once.

Yet another possibility (depending on how much data you have) might be time series analysis.

Answer 2

@Peter

Thank you for replying. Yes, we will be extremely cautious and not use any words in the direction of “cause” or “effect”. Believe me, we are very aware of this and I only used those words as I thought it simpler here. Our whole endeavor is very exploratory and we are looking into new things.

Our variable of interest is based on the following two questions; “Is there a shortage of suitable brides in your community?” and the follow-up question; “Is this a result of sex-selective abortions?”. From this we get a binary variable of whether or not there is such a shortage. Indeed, this is likely to be quite subjective in nature which brings its specific difficulties. The advantage is that with this variable we can analyze this on the micro-level which would be difficult with male-female ratios.

I don’t see how standard regressions would be suitable. This is why I wrote originally and why I am not sure what to do. Let me refer to the example with crime again and how it all would look in a standard regression. We are interested in how a shortage of brides as a result of sex-selective abortions (independent) is related to crime (dependent) but not what other factors (independent) are related to higher incidence of crime. We are, however, interested in how a shortage of brides as a result of sex-selective abortions (independent) is related to other variables of interest regardless of how they are related to crime or to any other variable we find is related with a shortage of brides.

KML