Syndicate content

Add new comment

Getting to better data: Do men say yes more often than women?

Markus Goldstein's picture

A couple of days ago, my wife and I were having one of the moments -- I was convinced we had had a detailed conversation about something and she was convinced that no such conversation had taken place.   Now, if you were to show up and do a survey of us, we wouldn't agree.

This disagreement between two people sharing significant aspects of their lives is the topic of an interesting 2001 paper by Kate Miller, Eliya Zulu, and Susan Watkins (quasi-gated version here).    While in the past, I have blogged about things people don't want their spouse to know, and the practical difficulties of getting husband and wife responses to the same question,  this paper provides a much more comprehensive overview of when husbands and wives might agree and when they might not.  

So let's start with why they might not agree.    In addition to secrets kept on purpose, the first, obvious, case is that reasonable people might not agree over what is actually going on.  Miller and co. give the example of a pit latrine that is falling apart.   She might say they have one, he might not.   Then there is the example of differing interpretations of what constitutes a discussion.  Miller and co. have an example I can relate to, where one partner is lacking substantive contributions and thus less likely to recall it (as a conversation anyhow).   Of course, issues such as moving into your husband's house can also play into this disagreement as well.   For example, a woman moving into her husband's house might not conceive of the assets as belonging to her household.      Finally, there is the problem of polygyny -- which Miller and co. solve by only analyzing data from monogamous couples.  

So Miller and co. set out to examine how large the difference in spousal responses are and what might be driving them by using data primarily from the Malawi Diffusion and Ideological Change Project (MDICP).   This is a neat dataset because they have husband and wife responses to whether or not the household has a range of household assets, livestock, a number of fertility questions and family planning and AIDS questions.    They use yes/no responses to make the agreement (or not) crystal clear for 17 different questions.    They also benchmark their results off of Malawi and Kenya DHS data.  

Before getting to what they find, they have a useful examination (and discussion of the literature) on whether or not it matters that a respondent is interviewed by an enumerator of the same sex.   According to the literature they cite, there is little evidence that this matters.   Indeed, in the MDICP, this is not part of the protocol, so they do some analysis and show that yes indeed, the matching of sex between enumerator and respondent doesn't matter.  
In terms of their main results the good news is that, in the majority of cases, husbands and wives agree as to what is going on.   In terms of whether or not the household has certain assets, this is pretty high -- the lowest level of agreement was on ownership of chickens and lamps which came in at 86 percent.   Children and fertility shows higher agreement, with the lowest agreement coming in on whether or not the wife was pregnant at 96 percent of the couples.   But family planning and AIDS questions show markedly lower agreement, with responses to whether or not the couple had discussed family size desires showing up with only 63 percent agreement.   The surprise for me was that men were more likely to say yes than women.   They benchmark these results against the DHS (where questions allow), and for the most part, these patterns are similar. 

Miller and co. go on to break down this disagreement into two pieces.   The first is the association effect -- which comes from the fact that they are a pair of people (if you are thinking of a 2X2 table of the spouses' responses, this is the data coming from within those 4 cells).   The second is the gender effect (and hence comes from the margins of that 2X2 table), i.e. the fact that men and women may systematically answer the questions differently.    They do some analysis to break out these different effects.   Overall, as one might expect, the association effect is fairly strong -- for example on the question on discussing family planning, one is about 5 times more likely to give the same response as one's spouse, rather than the opposite.   However, the association effect is weaker for family planning and AIDS questions than in other areas.   This pattern isn't evident for the gender effect -- men are more likely to say yes to most of the questions.   

Unpacking these results, and looking at regional differences, they come up with three interesting conclusions.   First, as noted above, men say yes more.   They can rule out enumerator error and ambiguity in the questions because these really shouldn't show a systematic pattern like this.  The explanation they settle on is that men are more likely to try and project themselves as good providers and good citizens in Malawi's efforts to curb population growth.  
The second pattern is that there is a greater discrepancy in husband and wife responses on family planning and AIDS questions.   There is no clear explanation for this, but what they can rule out is that this is because of covert birth control use by wives -- in these data husbands are more likely to report that the couple is currently using contraception than the wife is.   The third pattern brings in context.   Miller and co. find that the sole regional difference in spousal agreement is on family planning and AIDS -- which is evident in discordance for the South and Central, but not the North.     Laying out the different characteristics of these regions, Miller and co.  conclude that where women have higher status and autonomy (the North), there is more spousal agreement.  

So, this gives us food for thought as we design our surveys.   Clearly whom you ask matters.   And apparently, men are much more likely to say yes for things like assets, but also for behaviors like family planning.    Overall, as Miller and co. put it:  "the quality of the data should always be questioned."