I am currently working with survey data, and there are multiple questions in the survey that act as a measure of poverty.

One question asks how often the individual has had electricity in their residence, another is how much running water and there are about 5 other such questions, each measured on a scale 1 to 5 where 1 is rarely and 5 is all the time.

I want to create a dummy variable variable for living condition using those questions. How do I go about it? Does it make sense to just average them out for each observation or is there a more robust way. Another option might just be to run each of them separately, but I just wanted some opinions.

  • $\begingroup$ This question seems clear enough to me. I'm voting to leave open. $\endgroup$ – gung - Reinstate Monica Jan 24 '17 at 1:47
  • $\begingroup$ Note that when you create a combined variable, it doesn't have to be a dummy variable. $\endgroup$ – gung - Reinstate Monica Jan 24 '17 at 1:48

Some options other than averaging them out include principal component analysis, factor analysis and creating an index. Principal component analysis allows you to combine existing variables into new "dimensions" by exploring variance, and factor analysis similarly exploits patterns in how different variables vary (move) together. This World Bank note on poverty measurement says that you can create "naive" indices, but actually there's no reason why they have to be naive. Depending on what you want to find out from your data, you can aggregate your variables in sophisticated ways that reflect importance of individual variables, such as by assigning different weights or number of points to different variables (weighted average), combining them with conditional multipliers (i.e. if var A > 2, multiply/weigh the value of var B by X amount), etc.

You can learn more about creating indices by reading the methodologies (and their criticisms, lessons learned, and revisions) of existing indices like the Human Development Index or the Multidimensional Poverty Index.

Note that if you average the values of your ~7 variables for each individual, you won't get a dummy. You get a dummy out of each variable by assigning for example all values above 2 as "having" something (1) or "not having it" (0), or I guess you can have one dummy if people "have" more than 3 out of 7 things. However, I wouldn't reduce my variables to dummies unless* I thought that there's really no difference between people who answer 4 and 5 to one of the questions. And I wouldn't reduce them all to one dummy unless I thought not having water is the same as not having electricity. Think about it: if you believe answers 4 and 5 really reflect something meaningfully different, then you lose valuable information by collapsing them into one bit of data.

*Well, I'm lying; I would also probably reduce the number of responses if I were doing something where I depend on having a certain number of responses in each cell in a crosstable, like log-linear analysis - but that's a matter of analysis, not measurement.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.