Computing percentiles

[Ema Davies]

Hello,

I am trying to estimate percentile shares for income and wealth and do the histogram of wealth distribution by income. I tried using pshare command from Ben Jann that I found here:

pshare estimate wealth [aw=weight], p(10(10)90 95) pvar(income) density
pshare histogram, yline(1)

in order to obtain the figure as the one below:

However, pshare command does not work with aweights. Do you know of another way to produce the figure?

Thank you!

[Maarten Buis]

It is very very very rare that you need aweights. My prior is about 99.99999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 % (I could be off one or two 9s) that what you really want is pweights. In that somewhat likely case you can just use pshare

[Ema Davies]

It is very very very rare that you need aweights. My prior is about 99.99999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 99999999999999999999999999999999999999999999999999 % (I could be off one or two 9s) that what you really want is pweights. In that somewhat likely case you can just use pshare

I do actually - I am working with SCF+ (Survey of Consumer Finances) dataset that was constructed by Kuhn et al. (2020). I was going through their replication files and they also used aweights - I am trying to obtain some further results so I wanted to produce this histogram.

[Maarten Buis]

A quick look suggests to me that they should have used pweights

[Ema Davies]

A quick look suggests to me that they should have used pweights

Thank you. I am not sure what the right weights to use are. When constructing the data, the authors also did the imputation of missing variables and imputed five values for each missing observation (e.g. business_equity in the data example below). Given that each individual is then represented five times, is it still fine to use pweights or they may need to be rescaled depending on what I want to compute?

* Example generated by -dataex-. For more info, type help dataex
clear
input double year str9 id double wgtI95W95 float(income business_equity)
1949 "19490001"  .9987322315350646  55483.03 189821.06
1949 "19490001"  .9987322315350646  55483.03  10321.75
1949 "19490001"  .9987322315350646  55483.03  717922.2
1949 "19490001"  .9987322315350646  55483.03 17419.074
1949 "19490001"  .9987322315350646  55483.03  88968.34
1949 "19490002" 1.1133562129916736  42315.05         0
1949 "19490002" 1.1133562129916736  42315.05         0
1949 "19490002" 1.1133562129916736  42315.05         0
1949 "19490002" 1.1133562129916736  42315.05         0
1949 "19490002" 1.1133562129916736  42315.05         0
1949 "19490003" 1.1235737530466936  28666.23         0
1949 "19490003" 1.1235737530466936  28666.23         0
1949 "19490003" 1.1235737530466936  28666.23         0
1949 "19490003" 1.1235737530466936  28666.23         0
1949 "19490003" 1.1235737530466936  28666.23         0
end

[Maarten Buis]

That is a different problem. For that you need to use the mi commands, but I don't know how pshare interacts with mi

[Ema Davies]

That is a different problem. For that you need to use the mi commands, but I don't know how pshare interacts with mi

Thank you very much for your help!