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 |
Steve Samuels <sjsamuels@gmail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: AW: AW: st: error message "too many weights" |

Date |
Tue, 27 Mar 2012 16:03:37 -0400 |

Ah, SPSS! Base SPSS has no survey capabilities but does estimate "weighted" (i.e. frequency-weighted) statistics. I'll bet that many unsuspecting investigators have reported wrong standard errors and p-values for those statistics. You should use pweights. You need a very good reason to ignore the other survey design variables-PSU and stratum. Having "one sample" (a subgroup?) is not good reason unless its members reside in a very small number of PSUs. Then you would need to include the PSU & strata as covariates. Steve sjsamuels@gmail.com On Mar 27, 2012, at 1:43 PM, Jette Kellerhoff wrote: Thanks for your advice. It is true, that the data is part of a survey, but I have chosen only the subjects of one sample. The variable gew represents the real abundance pattern of these subjects - the values of gew are the expansion factors. The expansion factors of the subjects are multiplied by the probability to stay in the panel. Is it correct to use fweight in that case, or pweight? I am not used to Stata, normally I use SPSS. Jette -----Ursprüngliche Nachricht----- Von: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von Steve Samuels Gesendet: Dienstag, 27. März 2012 11:14 An: statalist@hsphsun2.harvard.edu Betreff: Re: AW: AW: st: error message "too many weights" Jette Kellerhoff- Your use of the words "panel" and "wave". suggests that you have survey data. If so, the weight should be designated a probability weight (pweight). At a minimum then, you need [pweight = gew] in the -stset- statement. This will give you properly weighted estimates of descriptive statistics like the survival curve. If the survey had a multi-stage design and you intend to run regression commands like -stcox-, you will need to -svyset- the data. If you don't, standard errors and p-values will be wrong. The -svyset- statement has the form: ********************************************* svyset psu_var [pw = gew], strata(stratum_var) ********************************************* After that, use the -svy:- prefix for -stcox- and -streg-. If you are planning multiple failure time analyses and weights are constant within id, then you will be omit the cluster(id) option in -stset-. That won't matter, because standard errors will be based on the survey design. If weights are not constant within ID, -stset- will complain. So check first. You have multiple spell data, apparently, and I'm not sure of your data structure. But a generic solution is to add an enter() option to the -stset- statement and omit the id() option. The code would be something like: *************CODE BEGINS************* svyset psu_var [pw = gew], strata(stratum_var) stset tfp , fail(event indicator) enter(start time for spell) exit(time.) svy: stcox // etc. ************************************** Steve sjsamuels@gmail.com See also: http://www.stata.com/statalist/archive/2008-10/msg01494.html. On Mar 26, 2012, at 5:03 AM, Nick Cox wrote: I could tell you how to force non-integers to integers but I fear that would be telling how to force your problem into an incorrect or dubious application of -st-. I am in no sense an expert on survival analysis, so I won't comment further. This may or may not be enough information for someone competent in that field to get a sense of what your real problem is and give better advice. Nick On Mon, Mar 26, 2012 at 9:44 AM, Jette Kellerhoff <jettek82@aol.com> wrote: > For each subject exists a weight (variable gew), which identifies the real > abundance pattern in time. The computed values of the variable gew are the > result of the real abundance pattern multiplied by the probability to stay > in the panel. So in many cases the results are not integers. How to create > integers? Nick Cox > I could tell you about Stata. But you really need to explain what the > variable -gew- is, why some values are not integers, and why you want > to treat them as frequency weights. On Sun, Mar 25, 2012 at 6:50 PM, Jette Kellerhoff <jettek82@aol.com> wrote: >> There are some non-integer values. How to create integers? Nick Cox >> The implication is that your variable -gew- contains non-integer >> values. Is that true? >> >> . list gew if gew != floor(gew) >> >> fweights indicate frequencies, which must be integers. On Sun, Mar 25, 2012 at 6:09 PM, Jette Kellerhoff <jettek82@aol.com> >>> I removed the first bracket, but the error message is still the same: may >>> not use noninteger frequency weights >>> stset tfp [fweight = gew], id(hhnrakt) f(des=2 4 6 8 10 12) exit(time .) Steve Samuels >>> You did not follow my advice to look at the examples. You only removed > the >>> last bracket; the beginning "[" of the second pair remains. On Mar 25, 2012, at 12:40 PM, Jette Kellerhoff wrote: >>> Thanks for your advice. If I remove the brackets, I get the following >> error >>> message may not use noninteger frequency weights. I don't understand it, >>> because fweights are allowed in the stset-command. >>> The do-file: >>> stset tfp [fweight=gew] [ ,id(hhnrakt) f(des=2 4 6 8 10 12) exit(time.) Steve Samuels >>> I should add that in the -help- for -stset-, the brackets indicate where >>> options, if any, should go. Before running an unfamiliar command, always >>> look at the examples at the bottom of the help screen or in the manual. >>> Remove the second pair of brackets in your -stset- statement. They are > not >>> legal Stata syntax. On Mar 22, 2012, at 3:18 PM, Jette Kellerhoff wrote: >>> Sorry for the impatientness, but I have to solve the problem very soon. >>> The case: >>> - 4841 subjects with multiple record per subject (max 4 records) >>> - for each subject exists a weight, which identifies the real abundance >>> pattern in time >>> >>> I used the following do-file: >>> gen des =tfin~=ti >>> >>> gen tf = tfin-tstart+1 >>> >>> gen org=1+2*(spell-1) >>> replace des=org >>> replace des=org+1 if tfin<ti >>> gen tfc=tstart if spell==1 >>> replace tfc=tfc[_n-1] if spell~=1 & hhnrakt==hhnrakt[_n-1] >>> gen tsp = tstart-tfc >>> gen tfp = tfin-tfc+1 >>> stset tfp [fweight=gew] [ ,id(hhnrakt) f(des=2 4 6 8 10 12) exit(time .)] Richard Herron >>> Jette, can you provide a reproducible example? That may help diagnose >>> your problem. Can you reproduce the error with -sysuse- data or >>> provide your console output, please? On Thu, Mar 22, 2012 at 14:34, Jette Kellerhoff <jettek82@aol.com> wrote: >>>> how to deal with weights according to survival data analysis? If I use >>>> fweights according to stset command, I get the error message "too many >>>> weights". Does anyone have any suggestions? * * 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/

**References**:**AW: AW: st: error message "too many weights"***From:*"Jette Kellerhoff" <jettek82@aol.com>

- Prev by Date:
**RE: st: RE: work with larger string length** - Next by Date:
**re: st: summarizing by more than one grouping variable** - Previous by thread:
**Re: st: wilcoxons rank sum in data set with three groups** - Next by thread:
**Re: AW: AW: st: error message "too many weights"** - Index(es):