st: Re: st: Re: st: Re: st: RE: Ratio of coefficients from two regressions and standard error‏.

[Steve Samuels <sjsamuels@gmail.com>]

You are welcome.  After re-reading's Maarten's post, I recommend that you use temporary scalars (via -tempname-) and variables throughout the program, not permanent ones. Here the -bootstrap- automatically takes care of the cross-equation correlations. Many people would use -suest- for doing this.  There are several advantages: 1) you can debug a complex program  before -bootstrapping it; 2)  you don't need to create scalars; 3) the equation names and coefficient names are used in the definition of the returned scalars, so there's less possibility of a mix-up.

 Here's the -suest- version.

*********************************
program myboot, rclass
    reg trunk length
    estimates store m1

    reg weight length
    estimates store m2

    suest m1 m2
    return scalar ratio =  ///
    [m2_mean]length/[m1_mean]length
end
***********************************


The disadvantage is that -suest- cannot handle commands, like -stcox- that do not create a score variable for each observation. 


On Dec 4, 2011, at 1:39 AM, meenakshi beri wrote:

Thanks a lot! I highly appreciate that.

Best,Meenakshi

________________________________
Meenakshi Beri
Graduate Teaching Assistant
Department of Economics
Wayne State University
dy5651@wayne.edu


----------------------------------------
> Subject: st: Re: st: Re: st: RE: Ratio of coefficients from two regressions and standard error‏.
> From: sjsamuels@gmail.com
> Date: Sun, 4 Dec 2011 00:45:18 -0500
> To: statalist@hsphsun2.harvard.edu
> 
> 
> 
> Maarten Buis diagnosed this problem at http://www.stata.com/statalist/archive/2007-10/msg00201.html The second bootstrap replicate refuses to run because the variable "uhatmain" created by your -predict- statement already exists in the data set; as a result only one replicate is created. Maarten's suggestion: use a temporary variable. Or, drop "uhatmain" at the end of the program.
> 
> Also drop all created scalars at the end of the program with a -scalar drop- statement; otherwise those from the last replicate will hang around.
> 
> 
> 
> Steve
> 
> 
> 
> 
> 
> On Dec 3, 2011, at 11:20 PM, meenakshi beri wrote:
> 
> Hello Steve,
> 
> Thanks for your reply and teaching me the way to estimate the ratio and standard error. I am using the same code (after modifying it as per panel data needs and multiple ratios needs) given by you, but I have been struggling with this insufficient observations error (otherwise my regressions are running fine without bootstrap). I have generated new identifier cluster variable also for the
> bootstrapped panels, as well as generated the new time variable also (whatever information I got using previously existing statalist answers for such a situation, I tried that but nothing worked). Is there something I am doing wrong?
> 
> I get the following error after bootstrap command:
> 
> insufficient observations to compute bootstrap standard errors
> no results will be saved
> r(2000);
> 
> 
> Here is my sample code:
> 
> 
> program my_xtboot, rclass
> xi: xtreg rwhappy hibp diab cancr l_inc_wealth hibp_wealth diab_wealth cancr_wealth hispanic if sample_black == 1, fe
> scalar define b1 = _b[hibp_wealth]
> scalar define b2 = _b[diab_wealth]
> scalar define b3 = _b[cancr_wealth]
> predict uhatmain, u
> 
> xi: xtreg uhatmain l_inc_wealth hispanic, be
> scalar define b8 = _b[l_inc_wealth]
> return scalar ratio1 = b1/b8
> return scalar ratio2 = b2/b8
> return scalar ratio3 = b3/b8
> end
> 
> *generating new identiers for bootstrapping
> 
> gen long newhhid = .
> replace newhhid = hhidpn
> 
> gen time2 = .
> replace time2 = 1 if year == 1992
> replace time2 = 2 if year == 1993
> replace time2 = 3 if year == 1994
> replace time2 = 4 if year == 1995
> 
> * declaring new panel and time variable
> tsset newhhid time2
> 
> * getting rid of missing values
> generate sample=1-missing(newhhid, time2, l_inc_wealth, rwhappy, hibp, diab, cancr, hispanic)
> keep if sample
> 
> *bootstrap ratio
> bootstrap ratio1 =r(ratio1) , reps(100)cluster(hhidpn) idcluster(newhhid) nowarn: my_xtboot if sample_black == 1
> estat bootstrap, all
> 
> And here is what I get:
> 
> bootstrap ratio1 =r(ratio1) , reps(100)cluster(hhidpn) idcluster(newhhid) nowarn: my_xtboot if sample_black == 1
> (running my_xtboot on estimation sample)
> : :
> Bootstrap replications (100)
> ----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
> xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 50
> xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 100
> insufficient observations to compute bootstrap standard errors
> no results will be saved
> r(2000);
> 
> 
> Best Regards,
> Meenakshi
> ________________________________
> Meenakshi Beri
> Graduate Teaching Assistant
> Department of Economics
> Wayne State University
> dy5651@wayne.edu
> 
> 
> ----------------------------------------
>> Subject: st: Re: st: RE: Ratio of coefficients from two regressions and standard error‏.
>> From: sjsamuels@gmail.com
>> Date: Sat, 3 Dec 2011 16:23:19 -0500
>> To: statalist@hsphsun2.harvard.edu
>> 
>> 
>> 
>> Bootstrapping the ratio should give good results with fewer assumptions about the regression coefficients.
>> 
>> Steve
>> *************CODE BEGINS*************
>> sysuse auto, clear
>> capture program drop myboot
>> 
>> program myboot, rclass
>> reg trunk length
>> scalar define b1 = _b[length]
>> 
>> reg weight length
>> scalar define b2 = _b[length]
>> 
>> return scalar ratio = b2/b1
>> end
>> 
>> bootstrap ratio =r(ratio) , reps(40): myboot
>> estat bootstrap, all
>> **************CODE ENDS**************
>> 
>> On Dec 2, 2011, at 10:44 AM, meenakshi beri wrote:
>> 
>> Thanks for your reply. One more question -- how to use Fieller's theorem and derive confidence limits using stata in this case?
>> 
>> Meenakshi Beri
>> Graduate Teaching Assistant
>> Department of Economics
>> Wayne State University
>> dy5651@wayne.edu
>> 
>> 
>> From: Paul.Silcocks@liverpool.ac.uk
>> To: statalist@hsphsun2.harvard.edu
>> Subject: st: RE: Ratio of coefficients from two regressions and standard error‏.
>> Date: Fri, 2 Dec 2011 09:23:04 +0000
>> 
>> On the assumption that the two regression coefficient estimates have a Normal distribution, their ratio would have a Cauchy distribution (with no defined variance) if their correlation is zero. If the correlation is non-zero the exact distribution is complicated, though under certain conditions it tends to a Normal distribution.
>> 
>> You'd be better off instead using Fieller's theorem to obtain confidence limits rather than estimating the standard error
>> 
>> Paul Silcocks BM BCh, MSc , FRCPath, FFPH, CStat
>> Senior statistician,
>> Cancer Research UK Liverpool Cancer Trials Unit
>> University of Liverpool
>> Block C Waterhouse Building
>> 1-3 Brownlow Street
>> L69 3GL
>> 
>> email: paul.silcocks@liverpool.ac.uk
>> tel: 0151 7948802
>> mob: 0794 983 2775
>> 
>> -----Original Message-----
>> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of meenakshi beri
>> Sent: 02 December 2011 06:08
>> To: statalist@hsphsun2.harvard.edu
>> Subject: st: Ratio of coefficients from two regressions and standard error‏.
>> 
>> Hello Statalist,
>> I am running a fixed effects regression followed by an auxiliary regression to capture the coefficient of time invariant variables. I want to estimate the ratio of two coefficients from these two regressions respectively along with the standard error of the ratio. How can I estimate the ratio and standard error?
>> Thanks,Meenakshi BeriWayne State University
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/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/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/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/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/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/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/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/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/