Hi, everyone. I am writing a paper using a difference-in-difference method to assess the impact of the 2009 WIC food package on increasing breastfeeding rates. (WIC stands for Special Supplemental Nutrition Program for Women, Infants, and Children.) I'm using National Immunization Survey data. My treatment group are women income-eligible for WIC (<185% federal poverty level), and the comparison group are women whose household income is just above 185% FPL.
My professor told me to test the parallel trends assumption using an event study analysis and to generate a plot using Stata's coefplot command. Here is the plot I generated for one of the variables (ever breastfed):
I have two questions:
First, how do I interpret this plot? I am assuming that each of the point estimates represents the difference in probability of ever breastfeeding between the treatment group and the comparison group for that year. Further, since the point estimates prior to year 2008 (the omitted year) all have confidence intervals that intersect with 0, there is no statistically significant difference between the two groups. This, in turn, would indicate that the data do not reflect any pre-trends, thus providing support for the parallel trends assumption. Is this correct? I will note that I've scoured the internet for a basic explanation of this type of graph, but I haven't found anything for a newbie like myself.
Second, my professor made an additional comment about there being some "year-on-year noise" in the results, but I am unclear about what that means. Can someone provide an explanation for that comment? (And, yes, I've asked for clarification from the professor, but so far no response.)
Would appreciate any thoughts a Stata user might have. I am happy to provide the code I used if that would be helpful.
Kelly Kogan
My professor told me to test the parallel trends assumption using an event study analysis and to generate a plot using Stata's coefplot command. Here is the plot I generated for one of the variables (ever breastfed):
I have two questions:
First, how do I interpret this plot? I am assuming that each of the point estimates represents the difference in probability of ever breastfeeding between the treatment group and the comparison group for that year. Further, since the point estimates prior to year 2008 (the omitted year) all have confidence intervals that intersect with 0, there is no statistically significant difference between the two groups. This, in turn, would indicate that the data do not reflect any pre-trends, thus providing support for the parallel trends assumption. Is this correct? I will note that I've scoured the internet for a basic explanation of this type of graph, but I haven't found anything for a newbie like myself.
Second, my professor made an additional comment about there being some "year-on-year noise" in the results, but I am unclear about what that means. Can someone provide an explanation for that comment? (And, yes, I've asked for clarification from the professor, but so far no response.)
Would appreciate any thoughts a Stata user might have. I am happy to provide the code I used if that would be helpful.
Kelly Kogan
Comment