I am working with a dataset composed of household surveys over 2 years. To correct for missing data, multiple imputation was used and five distinct iterations are provided for each respondent. There are 10,190 observations in my sample. I want to analyze this data in such a way that I can speak to potential implications of a federal relief program. The program was announced after the surveys were conducted, but the data included in the surveys allows me to identify households that are eligible for the relief. For those eligible households, I want to reduce household debt by the dollar amount prescribed in the relief program. I recognize that, because the surveys were conducted before the program's announcement, the responses of eligible households are from the perspective of existing debt levels.
I would like to recalculate an objective measure (e.g., debt ratio) of households after I have artificially introduced the relief program (i.e., by reducing household debt levels, for eligible households, by a specific dollar amount). I want to be able to draw comparisons to other, ineligible (untreated) households with similar debt ratios so that I can describe potential implications of this relief. For example, if pre-relief a household had a debt ratio of 65% and post-relief has a debt ratio of only 30%, I want to 'match' that household to other, similar household(s) with a similar debt ratio. I would like to be able to 'match' these households by more than one characteristic (e.g., education level, income, etc.).
Ultimately, I would like to be able to (in an empirically sound manner) suggest that these two households are quite similar across a vector of household demographic characteristics (education, income, etc.), with the exception of pre-relief household debt levels. My goal is to be able to draw comparisons between these matched households. After introducing debt relief and matching a treated household with an untreated household using the demographics described above, I then want to look at the differences in financial wellbeing of the treated and untreated household (e.g., retirement savings, use of financial planner, perception of financial stability) to observe discrepancies. My conclusion would be that, as a result of this program, we can expect eligible (treated) households to improve in these other financial areas by a specific amount (i.e., the 'gap' between the treated and untreated household financial characteristics).
My questions is - can I this, and if so, what method is most appropriate? I originally explored propensity score matching, but I'm not sure that's going to work given by 'artificial' treatment. I'm also curious about cluster analysis (using the -cluster- command), but I haven't used this approach before, and it seems to be appropriate only at the summary level. I'd very much appreciate suggestions and insight!
-Courtney
I would like to recalculate an objective measure (e.g., debt ratio) of households after I have artificially introduced the relief program (i.e., by reducing household debt levels, for eligible households, by a specific dollar amount). I want to be able to draw comparisons to other, ineligible (untreated) households with similar debt ratios so that I can describe potential implications of this relief. For example, if pre-relief a household had a debt ratio of 65% and post-relief has a debt ratio of only 30%, I want to 'match' that household to other, similar household(s) with a similar debt ratio. I would like to be able to 'match' these households by more than one characteristic (e.g., education level, income, etc.).
Ultimately, I would like to be able to (in an empirically sound manner) suggest that these two households are quite similar across a vector of household demographic characteristics (education, income, etc.), with the exception of pre-relief household debt levels. My goal is to be able to draw comparisons between these matched households. After introducing debt relief and matching a treated household with an untreated household using the demographics described above, I then want to look at the differences in financial wellbeing of the treated and untreated household (e.g., retirement savings, use of financial planner, perception of financial stability) to observe discrepancies. My conclusion would be that, as a result of this program, we can expect eligible (treated) households to improve in these other financial areas by a specific amount (i.e., the 'gap' between the treated and untreated household financial characteristics).
My questions is - can I this, and if so, what method is most appropriate? I originally explored propensity score matching, but I'm not sure that's going to work given by 'artificial' treatment. I'm also curious about cluster analysis (using the -cluster- command), but I haven't used this approach before, and it seems to be appropriate only at the summary level. I'd very much appreciate suggestions and insight!
-Courtney
