I am testing the effect of a product applied at four different concentrations (continuous variable), at two sites and six genotypes (class variables). The treatments were blocked at each site. Site*concentration is significant, which is interesting. Two specific follow-up questions are 1. what is the slope for concentration at each site, averaged across genotypes? and 2. are each of those slopes different from zero (i.e., does concentration actually affect the response at both of the sites)? This may not be the most elegant strategy, but I'm able to get the response = slope*concentration + intercept lines for each site by sending the lsmeans through proc reg. However, since the data variability is lost, this does not tell me whether each slope is different from zero. My attempts at proc mixed estimate statements for this test are giving errors; any help would be much appreciated! Thanks, Anna proc mixed data=data ;
class site genotype block;
model response = site|genotype|concentration ;
random block(site) ;
lsmeans site / at concentration = 0 ;
lsmeans site / at concentration = 10 ;
lsmeans site / at concentration = 20 ;
lsmeans site / at concentration = 30 ;
ods output lsmeans = lsmeans;
run;
proc sort data=lsmeans; by site;
proc reg data=lsmeans;
by site;
model estimate = concentration;
run;
... View more