Kayleigh de Bruin, s2594498 Tuesday, 9 February 2021
Chapter 10: Linear mixed modeling: introduction
10.2 Pre-post intervention designs.
Pre-test - data measured before the intervention.
Post-test - data measured after the intervention.
By representing a long format (table 10.2) we
acknowledge that there is really only one dependent
measure: in this case headache severity. The two
other variables indicate that this variable varies across both patients and time points (pre-
intervention and post-intervention).
Here we might consider applying a simple linear regression model, using a headache as
the dependent variable and measure (1st or 2nd) as a categorical predictor. However, since we
know that there is a correlation between the pre and post severity measures, we know that
measures also systematically vary across patients: some score high on average, and some score
low on average. The assumption of independence is therefore not tenable. Thus we have to run a
linear model, including not only an effect of the measure but also an effect of the patient. We
then have to decide between fixed effects or random effects for these variables.
Let’s first look at the variable measure. Since we are really interested in the effect of the
intervention, that is, we want to know how large the effect of aspirin is, we use a fixed effect for
the time effect (the variable measure). Moreover, the variable measure has only two levels,
which is another reason to opt for a fixed effect.
10.4: Reporting on a linear mixed model for pre-post data.
REML - Restricted Maximum Likelihood
ML - Maximum Likelihood
Video lecture
Chapter 10: Linear mixed modeling: introduction
10.2 Pre-post intervention designs.
Pre-test - data measured before the intervention.
Post-test - data measured after the intervention.
By representing a long format (table 10.2) we
acknowledge that there is really only one dependent
measure: in this case headache severity. The two
other variables indicate that this variable varies across both patients and time points (pre-
intervention and post-intervention).
Here we might consider applying a simple linear regression model, using a headache as
the dependent variable and measure (1st or 2nd) as a categorical predictor. However, since we
know that there is a correlation between the pre and post severity measures, we know that
measures also systematically vary across patients: some score high on average, and some score
low on average. The assumption of independence is therefore not tenable. Thus we have to run a
linear model, including not only an effect of the measure but also an effect of the patient. We
then have to decide between fixed effects or random effects for these variables.
Let’s first look at the variable measure. Since we are really interested in the effect of the
intervention, that is, we want to know how large the effect of aspirin is, we use a fixed effect for
the time effect (the variable measure). Moreover, the variable measure has only two levels,
which is another reason to opt for a fixed effect.
10.4: Reporting on a linear mixed model for pre-post data.
REML - Restricted Maximum Likelihood
ML - Maximum Likelihood
Video lecture
