Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

SV: SV: st: Control groups


From   <Alexander.Severinsen@telenor.com>
To   <statalist@hsphsun2.harvard.edu>
Subject   SV: SV: st: Control groups
Date   Tue, 13 May 2008 11:30:15 +0200

Steven, thanks a lot for the advice regarding sampsi, and for additional comments. I'll definitely look into it.

Best wishes,
Alex 

-----Opprinnelig melding-----
Fra: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] P vegne av Steven Samuels
Sendt: 13. mai 2008 05:13
Til: statalist@hsphsun2.harvard.edu
Emne: Re: SV: st: Control groups

Alexander, you can use -sampsi- as a rough guide for stratified sampling, provided that the relative sampling fractions (treatment/
control) do not differ much between strata.

I don't know this area, but do you really want to do a hypothesis test?  It sounds like you want to estimate proportions of people who respond to a campaign over time or at all. The control group provides the proportion who freely respond in the absence of a campaign. A test of zero campaign effect does not seem relevant.  Wouldn't the more pertinent questions be: how big an effect, answered with confidence intervals; or a null hypothesis that the effect is at most some quantity "D", with alternative effect>D.  The CI approach (but not a hypothesis test) can incorporate finite-population corrections.  See any sampling book for details.

-Steven


On May 12, 2008, at 4:10 PM, <Alexander.Severinsen@telenor.com>
<Alexander.Severinsen@telenor.com> wrote:

> Carlo, thanks for pointing me to sampsi!
>
> Steven, sorry about being sparse with information. I actually had many 
> different study designs in mind. Sometimes I will be using simple 
> random sampling, and don't intend to generalize my findings.
> Then all I am planning to do is testing whether proportions in treated 
> versus control are different. Also, my control group is internal.
>
> However, from time to time I would like to draw a stratified sample, 
> otherwise using the same approach as above. The way I understand you 
> sampsi would not be appropriate?
>
> Also, one particular study I will try to estimate is Lo (2002).  
> This is about the same as uplift modeling, uplift being another way of 
> saying "proportional hazards modelling". For this analysis I have come 
> across the Schoenfeld (1983), and the Stata program to estimate sample 
> sizes, stpower (findit stpower). Unfortunately, I don't have access to 
> Biometrics. So I am just guessing from the title that I am on the 
> right track!
>
> Lo, V.S.Y. (2002) "The True Lift Model - A Novel Data Mining Approach 
> to Response Modeling in Database Marketing." 4(2), p- 78-86.
>
> Schoenfeld, D. 1983.  Sample-size formula for the proportional-hazards
>         regression model.  Biometrics 39: 499-503.
>
> Best wishes,
> Alexander
>
>
> -----Opprinnelig melding-----
> Fra: owner-statalist@hsphsun2.harvard.edu [mailto:owner- 
> statalist@hsphsun2.harvard.edu] P vegne av Steven Samuels
> Sendt: 11. mai 2008 19:03
> Til: statalist@hsphsun2.harvard.edu
> Emne: Re: st: Control groups
>
> Alexander,
>
> Without more information, I cannot tell if -samnpsi- or any of the 
> other programs that you tried will give you proper answers.  They will 
> be okay, for example, if 1) you measure responses without sampling or 
> by simple random sampling; and 2) you don't intend to generalize your 
> findings beyond the two particular populations you study. Different 
> designs other than aimple two-group cross- sectional comparison might 
> reduce the needed sample size and strengthen your conclusions.  
> Whether your control group is internal (same
> population) or external (another population) matters, too.
>
> In any case, more details would be helpful.
>
> Steven
>
> On May 9, 2008, at 5:52 PM, <Alexander.Severinsen@telenor.com>
> <Alexander.Severinsen@telenor.com> wrote:
>
>> Dear Statalisters,
>>
>> Say I have a population of 500 000. I would like to treat this 
>> population with some sort of communication, and to be able to measure 
>> the effect of this treatment I have the opportunity to draw a control 
>> group. Based on earlier experiences the effects between the treated 
>> and the controlgroup could be as small as 0.5%.
>>
>> I want to be able to detect such a small effects, and I am wondering 
>> how large my control groups ideally would be to track these changes, 
>> say at an alpha level of 0.05.
>>
>> I have tried to use the fpower (findit fpower) and the simpower 
>> (findit
>> simpower) to determine optimal control group sizes. I am curious 
>> whether there are other alternatives in Stata?
>>
>> Thanks! And have a nice weekend.
>>
>> Best wishes,
>> Alexander
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/support/faqs/res/findit.html
>> *   http://www.stata.com/support/statalist/faq
>> *   http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index