1

I am somewhat confused about how best to interpret the results of my logistic mixed effects model.

I have two variables, confidence (continuous, 0-100) and meta-d', which can range from -0.5-2. From these two variables, I am trying to predict a binary variable, information seeking), whilst controlling for the effect of participant and trial number.

Thus the resultant model is:

lmer(information seeking ~ confidence * meta-d' + (1|Participant) + (1|Trial_n))

                                           estimate:,   std.error, df,        t,      p
inital_confidence                          -1.089e-02  6.304e-04  4.078e+03 -17.279  < 2e-16 
meta_d                                     -3.870e-01  1.319e-01  1.682e+02  -2.934  0.00381   
inital_confidence:meta_d                    4.390e-03  1.465e-03  4.073e+03   2.996  0.00275

I am confused about how to interpret the latter two results because of the way in which meta-d varies in the model. Any help or advice would be greatly appreciated - Thanks!

Rupert Riddle
  • 21
  • 2
    Does this answer your question? How to interpret interaction continuous variables in logistic regression? The coefficient for meta_d is the change in log-odds of your outcome per unit change in meta_d, when inital_confidence is at 0. The interaction coefficient is the extra change in that association between log-odds and meta_d per unit change in inital_confidence above 0. – EdM Aug 09 '22 at 20:39
  • I tried to follow the calculations posted in the answer to the linked post, but my numbers seem off. Would it be possible to check my working. (Please excuse the weird formatting i cant get it to work)
    When meta-d is 0 a unit change in confidence is associated with a (exp(-0.019)−1)≈ -0.02 decrease in the odds of information seeking.

    For every unit increase in meta-d' this effect of gonad increases by (exp(0.0039)−1)≈0.004%.

     – Rupert Riddle Aug 10 '22 at 10:21
  • Based on the above, how could i show the the predicted difference between whether information seeking is true or not given different levels of confidence and Meta-d. (Sorry if this is really obvious but im not getting it) – Rupert Riddle Aug 10 '22 at 10:23
  • With interactions you need to use tools for post-modeling evaluation of specific scenarios of interest, as the association of a predictor with outcome depends on the values of its interacting predictors. For that you need to combine the covariate values of interest with both the point estimates of log-odds coefficients and the covariance matrix of those estimates, while correcting for multiple comparisons. Many people find the R emmeans package helpful for that. – EdM Aug 10 '22 at 15:55

0 Answers0