0

I am trying a mixed model in R (lme4). The objective is to evaluate the average glucose level of patients in a hospital is decreasing in the last four years. The model is simple that glucose values are predicted from year of glucose measurement (year1, year2, year3 and year4) (year), number diabetes education appointment taken by the patient in the year (edu_appt) and time in months between the first appointment of the patient with the clinic and date of lab test done (months). Interaction term was introduced between edu_appt and the months to find out if there is any effect of education appointments in the long run..

Following is the result

Fixed effects:
              Estimate Std. Error         df t value Pr(>|t|)    
(Intercept)      8.167e+00  5.692e-02  1.292e+03 143.488  < 2e-16 ***
year2second      2.390e-01  4.921e-02  7.176e+03   4.857 1.22e-06 ***
year3third       5.843e-01  8.192e-02  7.164e+03   7.133 1.08e-12 ***
year4fourth      8.654e-01  1.153e-01  7.164e+03   7.506 6.84e-14 ***
edu_appt         4.966e-02  6.616e-03  7.630e+03   7.505 6.84e-14 ***
months          -2.298e-02  3.040e-03  7.168e+03  -7.560 4.51e-14 ***
edu_appt:months -1.402e-03  2.409e-04  7.304e+03  -5.819 6.16e-09 ***

Here if you notice, the estimate of edu_appt is possitve but when I add the interaction term months which is the duration between first_appt and hba1c measuremnt, It turned out to be negative. Can someone please tell me how do interpret this model.

arshad
  • 751
  • 4
  • 13
  • "How to interpret" questions tend to mean "teach me about this topic." Very broad, and likely covered in previous threads on this site. Suppose you read through them and then see whether you still have a specific question about your situation. – rolando2 Jan 17 '16 at 12:30
  • If you add an interaction between first_appt and hbalc then you should add both of those as main effects. – Peter Flom Jan 17 '16 at 12:46
  • @PeterFlom, they do have the main effects in their model ... – Ben Bolker Jan 18 '16 at 03:49
  • this is indeed a very general question (not at all specific to mixed models), but to my mild surprise I haven't been able to find a really good canonical answer on CrossValidated. [this](http://stats.stackexchange.com/questions/121863/citation-for-iv-coefficient-sign-change-with-inclusion-of-interaction-term) or [this](http://stats.stackexchange.com/questions/5450/what-if-interaction-wipes-out-my-direct-effects-in-regression) are good places to start. – Ben Bolker Jan 18 '16 at 03:58
  • @BenBolker Where are they? I am looking at the list and I don't see either first_appt or hbalc. – Peter Flom Jan 18 '16 at 11:58
  • my mistake, sorry. – Ben Bolker Jan 18 '16 at 13:24
  • by the way, this looks like output from `nlme::lme`, not `lme4::lmer` (not that it's very relevant) – Ben Bolker Jan 18 '16 at 13:26

0 Answers0