Skip to content
Prev 3532 / 20628 Next

Very weird lmer results, compared to SAS proc mix

On Sat, Mar 27, 2010 at 11:25 PM, Yong Wu <wuyong88 at gmail.com> wrote:
Both models are an unusual specification.  You have a random effect
with respect to age but no fixed effect for age.  This means that the
mean slope with respect to age across all ID's is constrained to be
zero, which for BMI is unlikely.  I think your model should be

BMI ~ 1 + exposure + age + (1 + age|ID)

I might even start with

BMI ~ 1 + exposure*age + (1 + age|ID)

unless exposure is at age zero.