08-03-2016 03:03 AM - edited 08-05-2016 12:18 PM
i have run a surveylogistic model and i got the point estimates:<0.001 , 95% Wald Confidence Limits:<0.001 <0.001.
except small sample size is there a another reason(mean)?
and can i use the result?
08-03-2016 06:33 AM
Please show the actual results and which procedure you were using. I don't think you actually had a confidence limit expressed as <0.001.
Generally the most common place I see SAS results like <0.0001 are associated with p-values. Which would mean that the for any practical test the result is statistically significant.
Without knowing exactly what your were doing it is hard to make a suggestion about "can I use the result" or not.
08-03-2016 08:34 AM - edited 08-03-2016 08:35 AM
proc surveylogistic data=hn NOMCAR; strata kstrata; cluster psu; weight wt;
domain subg; class sccatt(param=ref ref='1') sex(param=ref ref='1') obe_g(param=ref ref='1') sm_g(param=ref ref='1') alccat(param=ref ref='1') mas(param=ref ref='1') incm_g(param=ref ref='1') educ_g(param=ref ref='1') pmetcat(param=ref ref='1') dept(param=ref ref='1') hs(param=ref ref='1') urban(param=ref ref='1') sleepdur(param=ref ref='1');
model dp(event="1")=sccatt age sex obe_g sm_g alccat mas incm_g educ_g pmetcat dept fq_en hs urban sleepdur/rsq vadjust=none CLPARM;
08-03-2016 06:41 AM - edited 08-03-2016 06:43 AM
Just as SAS uses a special PVALUE format for p-values, there is a format for odds ratios to handle the case when one of the odds is almost zero or infinity. An odds ratio is computed as a fraction. A value that displays as <0.001 means that either the numerator is tiny or the denominator is huge.
Similalry, you might see >9999 if the numerator is huge or the denominator is tiny.
If you ever see a 95% expressed as (<0.001, >9999) then the analysis is telling you almost nothing. The true value of the parameter could be anywhere.