Date:         Fri, 30 May 2008 20:48:42 -0400
Reply-To:     Peter Flom <peterflomconsulting@mindspring.com>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject:      Re: modelling lactation curves
Comments: To: nchapinal@YAHOO.COM
Content-Type: text/plain; charset=UTF-8

>
>milk production follows a curve, as some of you may already now. When
>a high producer cow calves, the milk production per day is aroung 10
>kg a day, but increases very quickly getting soon a peak (even more
>than 40 kg/day) and then it starts decreasing slowly.
>I have seen paper that compare the milk production between two
>treatments by using PROC MIXED and REPEATED.
>
>For instance:
>
>proc mixed;
>class cow treatment;
>model production = treatment day treatment*day;
>repeated /subject=cow type=cs; *they usually try different types to
>see which one fits better;
>run;
>
>However, to me, that is not totally sound, since milk production is
>not linear.
>Is there any way to overcome this problem? Is it going to make any
>difference in the results?
>


Paul made the excellent suggestion of centering and then fitting a quadratic.   However, even this may fit your curve adequately.  With independent data, I'd suggest a spline, but, AFAIK, this isn't available in MIXED.  You might be able to do something in NLMIXED (and Dale may be chiming in with much better advice), but how many time points do you typically have?  If it's more than (say) 20 per cow, have you considered some form of time series analysis?

Peter

Peter L. Flom, PhD
Statistical Consultant
www DOT peterflom DOT com