[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
Jeph Herrin <junk@spandrel.net> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: Difference in Difference for Proportions |

Date |
Thu, 17 Sep 2009 16:53:47 -0400 |

Not sure whether this helps you, but I would normally test this with an interaction term in a model. For instance gen txsouth=t*south binreg f t south txsouth, n(pop) Then testing the coefficient on -txsouth- is the same as testing whether there is a significant difference in differences. hth, Jeph Misha Spisok wrote:

Hello, Statalist, In brief, how does one test a difference in difference of proportions? My question is re-stated briefly at the end with reference to the variables I present. A formula and/or reference would be appreciated if no command exists. I would like to test a difference in difference of proportions. -prtest- and -prtesti- do not work (easily) for my data, even for a simple test of differences. I have data grouped such that, for N states, I have the number of persons in state i with a condition (the variable for that count is f) and the population of state i in year y (pop). A "treatment" is applied and the pre-treatment period is t=0 and the post-treatment period is t=1. One can consider south=1 to be the treated and south=0 to be the non-treated group. For example, some observations may look like this: state year pop f t south 1 1990 1200000 10000 0 0 ... 50 1990 3000000 900 0 1 ... 1 2000 1500000 21000 1 0 ... 50 2000 3900000 2900 1 1 For differences in proportions within, for example, the pre-treatment period, for states in two regions (south==0 and south==1), I use, egen f_north_0 = sum(f) if south==0 & t==0 egen pop_north_0 = sum(pop) if south==0 & t==0 egen f_south_0 = sum(f) if south==1 & t==0 egen pop_south_0 = sum(pop) if south==1 & t==0 gen phat_n_0 = f_north_0/pop_north_0 /* proportion in north pre-treatment */ gen phat_s_0 = f_south_0/pop_south_0 /* proportion in south pre-treatment */ gen sp_n_0 = sqrt(phat_n_0*(1 - phat_n_0)/pop_north_0) /* standard error for phat_n_0 */ gen sp_s_0 = sqrt(phat_s_0*(1 - phat_s_0)/pop_south_0) /* standard error for phat_s_0 */ egen fn_0 = mean(f_north_0) egen fs_0 = mean(f_south_0) egen pn_0 = mean(pop_north_0) egen ps_0 = mean(pop_south_0) gen phat_0 = (fn_0 + fs_0)/(pn_0 + ps_0) /* pooled proportion, pre-treatment */ gen qhat_0 = 1 - phat_0 gen sp_0 = sqrt(phat_0*qhat_0*(1/pn_0 + 1/ps_0)) /* standard error of difference of proportions */ gen z_0 = (fs_0/ps_0 - fn_0/pn_0)/sp_0 (At this point I suppose I could use -prtesti- by summarizing the relevant variables then typing the results into the prtesti command...In any case, I think that neither -prtest- nor -prtesti- will help me with testing a difference in differences.) This, it would seem, allows me to test the difference in proportions in the pre-treatment period. Similarly, if I generate similar values for the post-treatment period, I can test the difference in proportions in the post-treatment period. egen f_north_1 = sum(f) if south==0 & t==1 egen pop_north_1 = sum(pop) if south==0 & t==1 egen f_south_1 = sum(f) if south==1 & t==1 egen pop_south_1 = sum(pop) if south==1 & t==1 gen phat_n_1 = f_north_1/pop_north_1 gen phat_s_1 = f_south_1/pop_south_1 gen sp_n_1 = sqrt(phat_n_1*(1 - phat_n_1)/pop_north_1) gen sp_s_1 = sqrt(phat_s_1*(1 - phat_s_1)/pop_south_1) egen fn_1 = mean(f_north_1) egen fs_1 = mean(f_south_1) egen pn_1 = mean(pop_north_1) egen ps_1 = mean(pop_south_1) gen phat_1 = (fn_1 + fs_1)/(pn_1 + ps_1) gen qhat_1 = 1 - phat_1 gen sp_1 = sqrt(phat_1*qhat_1*(1/pn_1 + 1/ps_1)) gen z_1 = (fs_1/ps_1 - fn_1/pn_1)/sp_1 How can I test (p_hat_s_1 - p_hat_s_0) - (p_hat_n_1 - p_hat_n_0), given that p_hat_* is a proportion? My uninformed guess is that it might be ((p_hat_s_1 - p_hat_s_0) - (p_hat_n_1 - p_hat_n_0)) / s, where s = some weighted version of sp_0 and sp_1. Many thanks, Misha * * 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/

**Follow-Ups**:**Re: st: Difference in Difference for Proportions***From:*Austin Nichols <austinnichols@gmail.com>

**References**:**st: Difference in Difference for Proportions***From:*Misha Spisok <misha.spisok@gmail.com>

- Prev by Date:
**st: Difference in Difference for Proportions** - Next by Date:
**st: AW: -areg- question** - Previous by thread:
**st: Difference in Difference for Proportions** - Next by thread:
**Re: st: Difference in Difference for Proportions** - Index(es):

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