4

So I'm not too familiar with mixed models, but wondering about the following:

library(reshape2)
library(lme4)
library(lmerTest)

#Simulate longitudinal data
N    <- 25
t    <- 2
x    <- rep(1:t,N)

#task1
beta1 <- 4
e1    <- rnorm(N*t, mean = 0, sd = 1.5)
y1    <- 1 + x * beta1 + e1

data1 <- data.frame(id=factor(rep(1:N, each=t)), day = x, y = y1, task=rep(c("task1"),length(y1)))
fakescore <- runif(50)
data1 <- cbind(data1, fakescore)
#temp <- reshape(data1, idvar=c("id"), varying=list(c(2,3),c(4,5), c(6,7), c(8,9), c(10,11), c(12,13),c(14,15)), v.names = varnames.lme, direction = "long")
model <-  lmer(fakescore ~ y*day + (1|id), data = data1)
summary(model)

The above is an example of how I generated my model. however, my real results are below. pp_aml, bmi, qsd, and time are continous (time is 0 and another timepoint). sex is male/female. But, I don't understand - what does it mean that qsd is significant and that qsd:time is at a trend level. I would like to know if qsd trajectory mirrors pp_aml trajectory but also want to know what is signficant down there, if somebody can explain the output to me? My actual stats knowlegde is limited so any help would be great.

Thanks!

model <-  lmer(pp_aml ~ bmi +sex+ qsd*time + (1|ID), data = temp)
Fixed effects:
              Estimate Std. Error         df t value Pr(>|t|)    
(Intercept)  1.920e+00  1.133e-01  3.743e+01  16.949  < 2e-16 ***
bmi          4.529e-03  3.676e-03  3.272e+01   1.232 0.226708    
sex         -1.851e-01  4.863e-02  3.184e+01  -3.808 0.000603 ***
qsd          1.483e-02  5.556e-03  6.595e+01   2.669 0.009573 ** 
time         6.535e-04  2.253e-04  3.690e+01   2.901 0.006237 ** 
qsd:time    -8.642e-05  4.284e-05  3.576e+01  -2.017 0.051229 .  
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Correlation of Fixed Effects:
         (Intr) bmi    sex    qsd    time  
bmi      -0.890                            
sex      -0.422  0.097                     
qsd      -0.438  0.205  0.118              
time     -0.189  0.022 -0.020  0.516       
qsd:time  0.064  0.039  0.011 -0.439 -0.701
  • 1
    These two previous answers might help you (it's got to do with interpreting main effects and interactions in regression): [interpreting main effects in the presence of an interaction](https://stats.stackexchange.com/questions/91074/the-main-effect-will-be-non-significant-if-the-interaction-is-significant/91095#91095), [how the meaning of 0 affects the interpretation of main effects and interactions](https://stats.stackexchange.com/questions/65898/why-could-centering-independent-variables-change-the-main-effects-with-moderatio/65911#65911) – Patrick Coulombe Jul 25 '20 at 03:15

1 Answers1

2

This question is about interpreting main effects and interactions.

The principle is that:

  • main effects that are not involved in an interaction can be interpreted as the association between a 1 unit change in that variable and the outcome, leaving all other fixed effects constant.

  • main effects that are involved in an interaction can be intepreted as the association between a 1 unit change in that variable and the outcome, when the other variable that it is interacted with is zero (or at it's reference level in the case of a categorical variable), and with other fixed effects held constant.

  • the interaction can be interpreted as the association between a 1 unit change in one variable and the outcome, when the other variable also changes by 1 unit.

Try not to be too bothered by statistical significance - it is practical significance that matters. The interaction term in your model is 3 orders of magnitude less than one of the main effects and 1 order of magnitude less than the other.

Robert Long
  • 53,316
  • 10
  • 84
  • 148
  • Does this answer your question ? If so please consider marking it as the accepted answer. If not please let us know why so that it can be improved – Robert Long Aug 07 '20 at 05:25