Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

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

From |
"Delahanty, Ryan" <ryan.delahanty@vanderbilt.edu> |

To |
"statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu> |

Subject |
st: logistic regression to assume equal number of observations |

Date |
Tue, 19 Jul 2011 05:33:11 -0500 |

I have a case-control (cc) dataset of some 10,000 people (5,000 cases and 5,000 controls). Each person has experienced up to 10,000 possible events. Each event can have a state 0-4. In addition, these people were collected at one of two different sites. Some people have experienced more events than others (total_events). Imagine the data takes this form: Person cc event event_state site total_events 1 1 1 1 1 3 1 1 10 2 1 3 1 1 100 1 1 3 2 0 10 4 0 2 2 0 1000 3 0 2 3 1 100 1 1 1 Unique lines do not exist for all person-event combinations, just for instances where a person experienced the event (i.e. event_state!=0). I'm trying to carry out a regression of the using -logistic- (Stata/SE 11.0 Win 64-bit) using the following form in a do-file: xi, prefix(): logistic cc i.event_state site total_events if event==1 xi, prefix(): logistic cc i.event_state site total_events if event==2 ... xi, prefix(): logistic cc i.event_state site total_events if event==10,000 For the most common events, I get a number of observations near 10,000 (i.e. everyone has experienced the event), but for most regressions my numbers will be between 2 and 10,000. My problem is that because having 10,000 entries for each possible individual-event combination would be prohibitively large (100M lines), I need to find an alternate way to run the above regression that will give me an identical result to what I would get if I ran a file of the size described, so that each regression has 10,000 observations. So for person 2 above, even though the event_state for event 1 would be 0, I want them included in the regression above (as with all people who would have event states of 0). How can I do this without including lines for all instances of event_state=0? Ideas about restructuring the data are also welcome so long as the equivalent regression could be run and the file does not become prohibitively large. Sorry I'm not familiar with a common dataset to help better illustrate this point. Any help is appreciated. * * 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/

- Prev by Date:
**Re: st: Two factor anova post test** - Next by Date:
**st: R: conditional probability** - Previous by thread:
**st: Problem with cross sectional regression, different results from xtreg, xi** - Next by thread:
**st: R: conditional probability** - Index(es):