1). Fit the liner model with GLM, no class statement.

model x = trt;

2 ) figure out what needs to be modified for estimate statement to estimate the slope.

estimate 'linear' trt -2 -1 0 1 2 / divisor=10;

If I was smart I would not need step 1.

On 8/12/08, cooch17@nospamverizon.net <cooch17@nospamverizon.net> wrote: > Suppose I have a single-classification ANOVA, where the sample means > show ordinal trend among successive levels. I construct a CONTRAST to > test for trend, in the usual manner. Fine - but I'm puzzled how to > interpret the parameter output from ESTIMATE (for the same contrast). > > For example, consider the following code snippet: > > data test; > > do trt=1 to 5; > > do i=1 to 5000; > > x=normal(0)*1.25+trt; > > output; > > end; > > end; > > > > proc glm; > > class trt; > > model x=trt / ss3; > > lsmeans trt; > > contrast 'linear' trt 2 1 0 -1 -2; > > estimate 'linear' trt 2 1 0 -1 -2; > > estimate '1 v 2' trt 1 -1 0 0 0; > > estimate '2 v 3' trt 0 1 -1 0 0; > > estimate '3 v 4' trt 0 0 1 -1 0; > > estimate '4 v 5' trt 0 0 0 1 -1; > > run; > > So, 5 treatment levels, linear increase of the treatment mean by 1 for > each level of trt. So, in fact, the 'slope' of the increase in the > treatment effect is 1 (confirmed by each of the paired estimate > statement contrasting successive treatment means). However, the output > from the linear ESTIMATE is (for a given simulated sample) something like > > estimate: -9.9968 > SE: 0.056 > > So, rounding off the estimate to -10 (which is what it should be... > > (2)(1)+(1)(2)+(0)(3)+(-1)(4)+(-2)(5) = 10 > > how do I interpret it? It clearly isn't the slope of the change in > treatment effect (even if I take it and divide it by 5 - the number of > levels of the treatment - it still isn't). While I *know* this is an > acceptable approach to testing significance of a trend, I also need to > estimate the slope of the trend. So, how to go from what ESTIMATE gives > me, to what I actually want/need. > > Pointers to the obvious? > > Thanks in advance. >