topic Re: How would you model this outcome? in Statistical Procedures

How would you model this outcome?

Melk — Mon, 07 Aug 2017 06:42:56 GMT

I want to compare the mean/median of a variable X between 2 independent samples. Variable X is continuous, ranges from <1 week to 32 weeks. The problem is, I have increments of weeks (1, 2, 3, 7, ...32) and then 30% of the data which is <1 week. How should I model this variable, should I assign it a randome value like 0.5 weeks?

Re: How would you model this outcome?

Rick_SAS — Mon, 07 Aug 2017 13:37:46 GMT

Are you counting <1 week as 0 weeks? If so, PROC FMM and PROC GENMOD can fit zero-inflated count models.such as ZIP and ZINB.

Re: How would you model this outcome?

StatDave — Wed, 09 Aug 2017 14:48:18 GMT

It probably doesn't matter what value you assign to the <1 cases if you use a rank-based nonparametric test as are available in PROC NPAR1WAY.

Re: How would you model this outcome?

Melk — Thu, 10 Aug 2017 19:41:50 GMT

Yes, this just occured to me too! Have just modelled it as 0 and used NP methods. Thank you all for your input.