RE: st: Losing Observations in Logit

[Isobel Williams <iwilliams24@hotmail.com>]

Fernando,

I've just tried to code my groups like that, and it works fine for groupA, but then when I replace groupB and groupC, Stata says '0 real changes made'. Do you know how to fix this?

Many thanks,
Isobel
----------------------------------------
> Date: Sat, 1 Mar 2014 09:21:31 -0500
> Subject: Re: st: Losing Observations in Logit
> From: f.rios.a@gmail.com
> To: statalist@hsphsun2.harvard.edu
>
> Yes, that is why i said your groups are not mutually exclusive. For
> instance, observations in zone 1 and cla_tam=1 are in group A and B.
> Groups in cla_tam are in all groups.
> Im not completely sure about the meaning of your variables, But making
> a blind shot, i think what you need is the following
>
> gen groupA=0
> gen groupB=0
> gen groupC=0
> replace groupA=1 if zona==1 & cla_tam==1 & h_selec==1
> replace groupB=1 if zona==1 & cla_tam==1 & h_selec==0
> replace groupC=1 if zona==0 & cla_tam==1 & h_selec==0
>
> Be sure that all observations are either in groups A B or C. If they
> are not. Im guessing you dont need to use those observations at all,
> but you should be able see that better since you know your data.
> Fernando
>
>
>
> On Sat, Mar 1, 2014 at 9:12 AM, Isobel Williams <iwilliams24@hotmail.com> wrote:
>> Fernando,
>>
>> Thanks a lot for your help- in just changing the code to groupA==1 | groupB==1, I now have 15,988 observations. However shouldn't I have 46,413 observations?
>> groupA= 32,187; groupB=14,226, so in a group A and group B sample I should have 32,187+14,226 =46,413
>>
>> I tried to make groups A, B, and C as exclusive as possible. Here is how my code, let me know what you think:
>>
>> //generate group A: in treatment area, eligible, and incorporated
>> generate groupA=1 if zona==1
>> replace groupA=1 if cla_tam==1
>> replace groupA=1 if h_selec==1
>> replace groupA=0 if zona==0
>> replace groupA=0 if cla_tam==2
>> replace groupA=0 if cla_tam==3
>> replace groupA=0 if h_selec==0
>> replace groupA=0 if h_selec==9
>>
>> ///generate group B: in treatment area, eligible, but not incorporated
>> generate groupB=1 if zona==1
>> replace groupB=1 if cla_tam==1
>> replace groupB=1 if h_selec==0
>> replace groupB=0 if zona==0
>> replace groupB=0 if cla_tam==2
>> replace groupB=0 if cla_tam==3
>> replace groupB=0 if h_selec==1
>> replace groupB=0 if h_selec==9
>>
>> ///generate group C: not in treatment area, eligible, not incorporated
>> generate groupC=1 if zona==0
>> replace groupC=1 if cla_tam==1
>> replace groupC=1 if h_selec==0
>> replace groupC=0 if zona==1
>> replace groupC=0 if cla_tam==2
>> replace groupC=0 if cla_tam==3
>> replace groupC=0 if h_selec==1
>> replace groupC=0 if h_selec==9
>>
>> Thanks,
>> Isobel Williams
>>
>> ----------------------------------------
>>> Date: Sat, 1 Mar 2014 08:57:10 -0500
>>> Subject: Re: st: Losing Observations in Logit
>>> From: f.rios.a@gmail.com
>>> To: statalist@hsphsun2.harvard.edu
>>>
>>> Hi Isobel,
>>> I think there are two main problems with your coding.
>>> First, your sample condition says:
>>> keep if groupA==1 & groupB==1
>>> But should say:
>>> keep if groupA==1 | groupB==1
>>> Since you want observations from both groups, (one or the other)
>>> The second problem I see, based on your previous mail, is that groups
>>> A, B and C should be mutually exclusive. Then again, perhaps the
>>> definition of treatment and groups are not connected.
>>> Fernando
>>>
>>>
>>>
>>> On Sat, Mar 1, 2014 at 8:48 AM, Isobel Williams <iwilliams24@hotmail.com> wrote:
>>>> Dear All,
>>>>
>>>> I am running a logistic regression, and then pairing them using propensity score matching. Within the dataset, I have divided the data into groups A, B, and C:
>>>>
>>>> generate groupA=1 if zona==1
>>>> replace groupA=1 if cla_tam==1
>>>> replace groupA=1 if h_selec==1
>>>> replace groupA=0 if zona==0
>>>> replace groupA=0 if cla_tam==2
>>>> replace groupA=0 if cla_tam==3
>>>> replace groupA=0 if h_selec==0
>>>> replace groupA=0 if h_selec==9
>>>>
>>>>
>>>> generate groupB=1 if zona==1
>>>> replace groupB=1 if cla_tam==1
>>>> replace groupB=1 if h_selec==0
>>>> replace groupB=0 if zona==0
>>>> replace groupB=0 if cla_tam==2
>>>> replace groupB=0 if cla_tam==3
>>>> replace groupB=0 if h_selec==1
>>>> replace groupB=0 if h_selec==9
>>>>
>>>> generate groupC=1 if zona==0
>>>> replace groupC=1 if cla_tam==1
>>>> replace groupC=1 if h_selec==0
>>>> replace groupC=0 if zona==1
>>>> replace groupC=0 if cla_tam==2
>>>> replace groupC=0 if cla_tam==3
>>>> replace groupC=0 if h_selec==1
>>>> replace groupC=0 if h_selec==9
>>>>
>>>> when I run the tab command for all groups, Stata tells me that:
>>>>
>>>> . tab groupA
>>>>
>>>> groupA | Freq. Percent Cum.
>>>> ------------+-----------------------------------
>>>> 0 | 45,316 58.47 58.47
>>>> 1 | 32,187 41.53 100.00
>>>> ------------+-----------------------------------
>>>> Total | 77,503 100.00
>>>>
>>>> . tab groupB
>>>>
>>>> groupB | Freq. Percent Cum.
>>>> ------------+-----------------------------------
>>>> 0 | 63,277 81.64 81.64
>>>> 1 | 14,226 18.36 100.00
>>>> ------------+-----------------------------------
>>>> Total | 77,503 100.00
>>>>
>>>>
>>>> . tab groupC
>>>>
>>>> groupC | Freq. Percent Cum.
>>>> ------------+-----------------------------------
>>>> 0 | 56,780 73.26 73.26
>>>> 1 | 20,723 26.74 100.00
>>>> ------------+-----------------------------------
>>>> Total | 77,503 100.00
>>>>
>>>>
>>>> However, when I run a logistic regression on propensity score matching between groups A and B, Stata tells me that I only have 7099 observations.
>>>>
>>>> Furthermore, when I wrote the code for keep if, the programme says "63227 observations deleted".
>>>>
>>>> Here is what I tried to do in estimating a logit propensity score and match (nearest neighbor) between groups A & B:
>>>>
>>>> preserve
>>>> keep if groupA==1 & groupB==1
>>>> logit treat floor fmiss wall hhinc2....
>>>> predict double ps1
>>>> psmatch2 treat, outcome (S06A20) pscore (ps1) caliper(0.2) common logit
>>>> restore
>>>>
>>>> The objective is to estimate a logit/propensity score and match observations from group A with group B. Any help on this matter would be very much appreciated.
>>>>
>>>> Many thanks,
>>>> Isobel Williams
>>>> *
>>>> * For searches and help try:
>>>> * http://www.stata.com/help.cgi?search
>>>> * http://www.stata.com/support/faqs/resources/statalist-faq/
>>>> * http://www.ats.ucla.edu/stat/stata/
>>> *
>>> * For searches and help try:
>>> * http://www.stata.com/help.cgi?search
>>> * http://www.stata.com/support/faqs/resources/statalist-faq/
>>> * http://www.ats.ucla.edu/stat/stata/
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/faqs/resources/statalist-faq/
>> * http://www.ats.ucla.edu/stat/stata/
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/faqs/resources/statalist-faq/
> * http://www.ats.ucla.edu/stat/stata/ 		 	   		  
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/