using distplot for categorical variables, pre and post matching

[Rose Matthews]

Hello all
I was recommended to use -distplot - to demonstrate the goodness of matching pre matching and post matching

As my amateur solution was initially this:
- However I had a problem how to overlay these graphs onto one

Prematching:

gen no = 1

graph bar sum(no) , over($treatment) over(Gender) asyvars


Post matching: 

gen new1 = _weight*10
gen newweight = int (new1)    //therefore controls with 0.33 become 0 

graph bar sum(newweight) , over($treatment) over(Gender) asyvars

However, someone told me to use -distplot-

I got to this:

gen new=_weight*10
gen new3=int(new)
distplot gender [fw=new3]

The graph doesn't make sense - it's perhaps because I don't understand how to use it properly.

1. Would you recommend using distplot, and can you advise how to use it
2. If you wouldn't advise at point 1 can you advise how to overlay both graphs together ?


Sample data:

* Example generated by -dataex-. For more info, type help dataex
clear
input float(gender smoking infection socialdeprivation ethnicity) double _pscore byte(_treated _support) double(_weight _infection) byte(_id _n1 _n2 _n3) float _nn double _pdif
1 0 0 7 4 .14629124373657137 0 1  .3333333333333333                 .  1 .  . . 0                    .
0 0 0 7 1 .18353598336623161 0 1  .3333333333333333                 .  2 .  . . 0                    .
1 0 0 6 4 .23642515629258887 0 1  .6666666666666666                 .  3 .  . . 0                    .
1 0 1 4 1 .27681680297995453 0 1  .3333333333333333                 .  4 .  . . 0                    .
0 0 0 6 1  .2888530710771261 0 1  .3333333333333333                 .  5 .  . . 0                    .
0 0 0 6 2  .3595669083790513 0 1  .3333333333333333                 .  6 .  . . 0                    .
1 0 0 3 1  .4088562831270797 0 1  .6666666666666666                 .  7 .  . . 0                    .
1 0 0 2 1  .5554993045631528 0 1 1.6666666666666665                 .  8 .  . . 0                    .
1 0 1 1 1  .6930733259357975 0 1 1.3333333333333333                 .  9 .  . . 0                    .
1 0 0 2 3  .7048182691423677 0 1                  1                 . 10 .  . . 0                    .
1 1 1 5 1 .17480944488523578 1 1                  1                 0 11 2  1 3 3  .008726538480995832
1 1 1 5 3  .2881297448796828 1 1                  1 .3333333333333333 12 5  4 3 3 .0007233261974433081
0 1 1 5 2  .5035916419173779 1 1                  1                 0 13 8  7 6 3  .051907662645774844
1 1 1 2 1  .5554993045631528 1 1                  1 .3333333333333333 14 8  9 7 3                    0
1 1 0 2 2  .6333538983172218 1 1                  1 .3333333333333333 15 9 10 8 3   .05971942761857574
1 1 0 2 2  .6333538983172218 1 1                  1 .3333333333333333 16 9 10 8 3   .05971942761857574
1 1 1 1 1  .6930733259357975 1 1                  1 .3333333333333333 17 9 10 8 3                    0
0 1 1 1 1  .8866625742391517 1 0                  .                 . 18 .  . . .                    .
0 1 1 1 1  .8866625742391517 1 0                  .                 . 19 .  . . .                    .
0 1 1 1 3  .9372933094185172 1 0                  .                 . 20 .  . . .                    .
end
label values gender Gender
label def Gender 0 "Female", modify
label def Gender 1 "Male", modify
label values smoking Smoking
label def Smoking 0 "Nonsmoker", modify
label def Smoking 1 "Smoker", modify
label values socialdeprivation social
label def social 1 "Most deprived", modify
label def social 7 "Least deprived", modify
label values ethnicity Ethnicity
label def Ethnicity 1 "White", modify
label def Ethnicity 2 "Asian", modify
label def Ethnicity 3 "Black African", modify
label def Ethnicity 4 "Mixed", modify
label values _treated _treated
label def _treated 0 "Untreated", modify
label def _treated 1 "Treated", modify
label values _support _support
label def _support 0 "Off support", modify
label def _support 1 "On support", modify

[Nick Cox]

I come into this thread as an expert on distplot, which is from the Stata Journal, and was written by me. Conversely, I don't know precisely what matching means here, but I can't see that even people who know enough to have a worthwhile opinion have been told what you've done.

Given gender, which for you is 0 for female and 1 for male, then distplot will show a jump in cumulative probability (1) from 0 to the probability of being female and (2) from that to the probability of being female or male, which is 1. That is not a helpful graph, but it does make sense.

distplot will show whether two or more distributions match in terms of having similar level, spread and shape, because if so they will have similar cumulative distribution functions. That makes most sense for measured variables or counted variables.

graph bar results can't usefully be overlaid on distplot results. This is a subtle point and not obvious, but distplot uses the machinery of twoway and isn't compatible with graph bar.

That's a long way from a complete or really helpful answer, but it's the best I can do.

[Rose Matthews]

Dear Nick,


yes I'm looking for this

'distplot will show whether two or more distributions match in terms of having similar level, spread and shape, because if so they will have similar cumulative distribution functions. That makes most sense for measured variables or counted variables'

Perhaps as I'm not able to overlay graphbar is the reason why -distplot- was recommended. Although to be honest I'm finding it quite difficult to use... so I may stick to my amateur version

Would you be kind enough to demonstrate sample code that plots two categorical variables pre and post matching?
Or perhaps as matching is not your specialisation amongst the many you have from what I've seen on this forum, I perhaps will wait for someone else.

[Nick Cox]

Thanks for this. You're right that you should get a better answer from someone else on matching. But if you want to compare distributions of categorical variables, you could do worse than dot or bar charts and even chi-square tests. All that said, don't the matching commands provide handles for assessing how well they work? I have no idea!

[Rose Matthews]

Maybe i'm just being pedantic...I have the stats.... I already have a loveplot... just determined to explore options with barcharts...