psmatch2 creating a table of before and after PSM

[Denise Vella]

Question: How do I construct a table showing the number of matches to assess covariate balance I.e no of matched treatments to number of controls?

Unfortunately, this is neither part of the psmatch2 or pstest output. Therefore one has
to construct the tables themselves.

David Radwin Melissa Garrido Oyvind Snilsberg

1. May I confirm this is correct? I have come up with the following code (using dummy data)

webuse cattaneo2
//Research question: Does alcohol affect bweight

**Creating macros
global treatment alcohol //intervention/treatment
global ylist bweight //outcome
global xlist mage medu msmoke //variables to match on t

//Ensuring replication
//Set seed to replicate the results

set seed 1234
gen sort_id = uniform()
sort sort_id

//Calculate psscore using psmatch2
psmatch2 $treatment $xlist, outcome($ylist) neighbor(3) bw (0.06) common logit

//Drawing graphs

pstest $xlist, treated($treatment) both

This is where I would appreciate your help
//Creating tables to show matched treatments to controls

//Drop the missing _weight values. If matched _weight = >0
drop if _weight == .

//Trying to summarize the number of matched treatment and controls

tabstat $treatment=1 [aw=_weight], by($xlist)

This returns an error with too many weights.

I understand that if _weight > 0 this is equal to a control that has been matched

_______
Also, what are your thoughts on Stephen Porter's website who quotes someone else that psmatch2 shouldn't be used. What are your thoughts about this? As I was rather happy with psmatch2
https://stephenporter.org/understand...atas-psmatch2/

[Andrew Musau]

psmatch2 is from SSC, as you are asked to explain (FAQ Advice #12). The matched sample is identified by nonmissing values of the generated "_weight" variable.

webuse cattaneo2, clear
psmatch2 mbsmoke mmarried c.mage##c.mage fbaby medu, out(bweight) ate logit
*FULL SAMPLE
bys _treated: summarize if e(sample), sep(0)
*MATCHED SAMPLE
bys _treated: summarize if !missing(_weight), sep(0)

Res.:

. *FULL SAMPLE

.
. bys _treated: summarize if e(sample), sep(0)

---------------------------------------------------------------------------------------------------------------------------------
-> _treated = Untreated

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
     bweight |      3,778    3412.912    570.6871        340       5500
    mmarried |      3,778    .7514558    .4322261          0          1
       mhisp |      3,778    .0362626    .1869675          0          1
       fhisp |      3,778    .0378507    .1908604          0          1
     foreign |      3,778      .05982    .2371845          0          1
     alcohol |      3,778     .018793    .1358113          0          1
    deadkids |      3,778    .2458973     .430675          0          1
        mage |      3,778    26.81048    5.645477         13         45
        medu |      3,778    12.92986    2.534403          0         17
        fage |      3,778    27.84436    8.794065          0         60
        fedu |      3,778     12.6739    3.481189          0         17
   nprenatal |      3,778    10.96294     3.51843          0         33
    monthslb |      3,778    21.89836    31.50073          0        272
       order |      3,778    1.859185    1.103474          0         11
      msmoke |      3,778           0           0          0          0
     mbsmoke |      3,778           0           0          0          0
       mrace |      3,778    .8478031    .3592592          0          1
       frace |      3,778    .8268925    .3783902          0          1
    prenatal |      3,778    1.177607    .4726644          0          3
  birthmonth |      3,778    6.513764    3.346774          1         12
    lbweight |      3,778    .0489677    .2158291          0          1
       fbaby |      3,778    .4531498    .4978661          0          1
   prenatal1 |      3,778    .8268925    .3783902          0          1
     _pscore |      3,778    .1704241    .1051475   .0134296   .7907192
    _treated |      3,778           0           0          0          0
    _support |      3,778           1           0          1          1
     _weight |        286    3.020979    3.123131          1         19
    _bweight |      3,778    3164.002      633.16        680       4734
         _id |      3,778      1889.5    1090.759          1       3778
         _n1 |      3,778    4027.722    224.3399       3779       4642
         _nn |      3,778           1           0          1          1
       _pdif |      3,778    .0003409    .0019572          0   .0424453

---------------------------------------------------------------------------------------------------------------------------------
-> _treated = Treated

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
     bweight |        864     3137.66    560.8931        397       5018
    mmarried |        864    .4733796      .49958          0          1
       mhisp |        864    .0243056    .1540853          0          1
       fhisp |        864    .0335648    .1802104          0          1
     foreign |        864     .025463    .1576177          0          1
     alcohol |        864    .0914352    .2883939          0          1
    deadkids |        864     .318287    .4660813          0          1
        mage |        864    25.16667    5.301348         14         43
        medu |        864    11.63889    2.167743          0         17
        fage |        864    24.74306    11.14795          0         55
        fedu |        864     10.7037    4.097048          0         17
   nprenatal |        864    9.862269     4.20762          0         40
    monthslb |        864    28.21991    36.92362          0        220
       order |        864     2.03588    1.182083          0         12
      msmoke |        864    2.146991    .7674824          1          3
     mbsmoke |        864           1           0          1          1
       mrace |        864    .8090278    .3932949          0          1
       frace |        864     .755787    .4298684          0          1
    prenatal |        864     1.30787    .6296153          0          3
  birthmonth |        864    6.655093    3.412406          1         12
    lbweight |        864    .1099537    .3130132          0          1
       fbaby |        864    .3715278     .483493          0          1
   prenatal1 |        864    .6898148    .4628372          0          1
     _pscore |        864    .2547889    .1304696   .0411911   .7889756
    _treated |        864           1           0          1          1
    _support |        864           1           0          1          1
     _weight |        288    13.11806    15.88986          1         80
    _bweight |        864    3334.843    626.2059        662       4933
         _id |        864      4210.5    249.5596       3779       4642
         _n1 |        864    2672.326    949.1435         38       3777
         _nn |        864           1           0          1          1
       _pdif |        864     .000148    .0009398          0   .0144565


.
. *MATCHED SAMPLE

.
. bys _treated: summarize if !missing(_weight), sep(0)

---------------------------------------------------------------------------------------------------------------------------------
-> _treated = Untreated

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
     bweight |        286    3368.675    614.8157        662       4933
    mmarried |        286     .506993    .5008274          0          1
       mhisp |        286    .0664336     .249475          0          1
       fhisp |        286    .0699301    .2554762          0          1
     foreign |        286    .0804196    .2724183          0          1
     alcohol |        286    .0104895    .1020583          0          1
    deadkids |        286    .2762238    .4479126          0          1
        mage |        286    25.84615    5.921484         14         43
        medu |        286     11.6049     3.29931          0         17
        fage |        286    25.62238    11.39178          0         60
        fedu |        286    11.16783    4.308272          0         17
   nprenatal |        286     10.0979    4.071837          0         25
    monthslb |        286    24.28671     35.5531          0        199
       order |        286    2.052448    1.332737          0          8
      msmoke |        286           0           0          0          0
     mbsmoke |        286           0           0          0          0
       mrace |        286    .7342657    .4424977          0          1
       frace |        286    .6923077    .4623475          0          1
    prenatal |        286     1.29021    .6238325          0          3
  birthmonth |        286    6.482517    3.415862          1         12
    lbweight |        286    .0594406    .2368619          0          1
       fbaby |        286    .4055944    .4918674          0          1
   prenatal1 |        286    .7062937    .4562574          0          1
     _pscore |        286    .2466459    .1456376   .0411911   .7889756
    _treated |        286           0           0          0          0
    _support |        286           1           0          1          1
     _weight |        286    3.020979    3.123131          1         19
    _bweight |        286    3177.224    577.7134        680       4734
         _id |        286    2508.112    1094.366         38       3777
         _n1 |        286    4180.755    269.7205       3779       4642
         _nn |        286           1           0          1          1
       _pdif |        286    .0001703    .0011881          0   .0144565

---------------------------------------------------------------------------------------------------------------------------------
-> _treated = Treated

    Variable |        Obs        Mean    Std. dev.       Min        Max
-------------+---------------------------------------------------------
     bweight |        288    3178.271    572.5312        680       4734
    mmarried |        288    .4965278    .5008582          0          1
       mhisp |        288    .0347222     .183394          0          1
       fhisp |        288    .0416667    .2001741          0          1
     foreign |        288      .03125    .1742955          0          1
     alcohol |        288      .09375     .291988          0          1
    deadkids |        288    .3090278     .462897          0          1
        mage |        288    25.84375    5.890987         14         43
        medu |        288    11.73264    2.929136          0         17
        fage |        288    25.19444    11.71007          0         55
        fedu |        288    10.51736    4.651594          0         17
   nprenatal |        288    9.659722    4.006377          0         21
    monthslb |        288    28.65278    41.24759          0        220
       order |        288    2.041667    1.240293          1         10
      msmoke |        288    2.107639    .7505241          1          3
     mbsmoke |        288           1           0          1          1
       mrace |        288    .7847222    .4117303          0          1
       frace |        288    .7430556    .4377091          0          1
    prenatal |        288    1.319444    .6374249          0          3
  birthmonth |        288    6.920139    3.268029          1         12
    lbweight |        288    .1006944    .3014475          0          1
       fbaby |        288      .40625    .4919872          0          1
   prenatal1 |        288    .6909722     .462897          0          1
     _pscore |        288    .2458135    .1451786   .0411911   .7889756
    _treated |        288           1           0          1          1
    _support |        288           1           0          1          1
     _weight |        288    13.11806    15.88986          1         80
    _bweight |        288    3367.649    618.2156        662       4933
         _id |        288    4179.389    268.6858       3779       4642
         _n1 |        288    2504.229    1095.575         38       3777
         _nn |        288           1           0          1          1
       _pdif |        288    .0001626    .0011807          0   .0144565


.

[Denise Vella]

psmatch2 is from SSC, as you are asked to explain (FAQ Advice #12). The matched sample is identified by nonmissing values of the generated "_weight" variable.


[/CODE]

Dear Andrew, a clarification required with this statement
So you say the matched sample is identified by 'non missing values of the generated _weight'

So why does psmatch2 generate different summary total values in the Treatment assignment table:

When compared with
drop if _weight == .

And if one for example looks at a particular variable, let's say:
tab medu _treated, column

The totals as you can see change (of course as the _weight == . dropped)




I assume when one proceeds to look at side-by-side boxplots, non -parametric density box plots, although one plots the results following
drop if _weight == .
Although the cumulative total changes, the shape of the graphs don't as these tests are not based on sample sizes

Is that intepretation correct?

[Andrew Musau]

I do not follow what you are asking. The treated units in the matched sample are a subsample of the treated units in the full dataset and the same for untreated units. Appending to the code in #2

assert !((_treated & !missing(_weight)) & (!_treated& e(sample)))
assert !((!_treated & !missing(_weight)) & (_treated& e(sample)))

we see that is the case. Res.:

. assert !((_treated & !missing(_weight)) & (!_treated& e(sample)))

. assert !((!_treated & !missing(_weight)) & (_treated& e(sample)))

.

So in my example, out of the 3778 untreated observations in the full sample, 286 are selected in the matched sample. For treated units, 288/864 observations are selected into the matched sample. For a large number of variables, a table should do for a comparison of means - unless you have some select variables of interest in which case you may consider simple graphs such as bar graphs.

[Denise Vella]

Yes. I understood this was a sample of the code. Perhaps what I was asking was in reference to the results from psmatch2.

i am trying to create a table to compare means +sd for continuous variables and number(%) for categorical with the SMD in the unmatched and matched sample.

if I could make the post more articulate, after using psmatch2, the results of the total treated & total untreated

ARE DIFFERENT

when compared to when won
1. performs psmatch2
2. Drop if _weight = . (Eg. 400 observations dropped)
3. tab medu _treated, column

The totals as you can see change (see screenshot above)

so with regards to categorical variables regarding n (%) for the table should I use the results from Step 3?

hope my question is not too complex

[Andrew Musau]

As I stated in #4, I can not follow your concern. Why would you expect that the totals will be the same before and after matching? I also do not know how detailed your understanding of the theory behind PSM is. If you look at the table that you attach in #1, you see that the matched sample is smaller than the full sample. As I stated in #4, it is a subsample of the full sample. What the table aims to show is that in terms of observables, the treated sample and control sample are closer in the matched sample compared to the full sample. I will now bail out of this thread, perhaps someone else can understand your concern which I fail to see.

[Denise Vella]

Yes. I understood this was a sample of the code. Perhaps what I was asking was in reference to the results from psmatch2.

i am trying to create a table to compare means +sd for continuous variables and number(%) for categorical with the SMD in the unmatched and matched sample.

if I could make the post more articulate, after using psmatch2, the results of the total treated & total untreated

ARE DIFFERENT

when compared to when won
1. performs psmatch2
2. Drop if _weight = . (Eg. 400 observations dropped)
3. tab medu _treated, column

The totals as you can see change (see screenshot above)

so with regards to categorical variables regarding n (%) for the table should I use the results from Step 3?

hope my question is not too complex