Kayleigh de Bruin, s2594498 Tuesday, 16 February 2021
Chapter 11.1: Pre-mid-post intervention designs
You have to choose whether you include a (categorical) variable either as a fixed effect or as a
random effect.
Rules of thumb:
→ use fixed for variables with a low number of categories if you are interested in exact effect
sizes (group means differences).
→ use random for variables with a large number of categories and are not interested in the
size of the effects.
→ use random if your categories represent random instances of a larger population of
categories.
→ a continuous variable is always treated as fixed.
Example
Research question: what is the mean difference in the measures on A, B, and C.
First, put it in a long format. Then, we have the categorical predictor variables measure,
child, and sex. How shall we treat the measure variable, fixed or random? → fixed effect as we
are interested in the differences between the measures.
Ability = b0 + b1 * measureA + b2 * measureB + …? + …. + e
E ~ N(0, ơe^2)
What shall we do with the child variable, random or fixed? → the children that we
observe here were randomly drawn from a population of children. We are not interested in
these particular children. So using dummy variables for each child and estimating the
differences seems silly. Therefore we use a random effect for the child differences.
Ability = b0 + b1 * measureA + b2 * measureB + childi + …. + eij
Eij ~ N(0, ơe^2)
Childi ~ N(0, ơe^2)
Chapter 11.1: Pre-mid-post intervention designs
You have to choose whether you include a (categorical) variable either as a fixed effect or as a
random effect.
Rules of thumb:
→ use fixed for variables with a low number of categories if you are interested in exact effect
sizes (group means differences).
→ use random for variables with a large number of categories and are not interested in the
size of the effects.
→ use random if your categories represent random instances of a larger population of
categories.
→ a continuous variable is always treated as fixed.
Example
Research question: what is the mean difference in the measures on A, B, and C.
First, put it in a long format. Then, we have the categorical predictor variables measure,
child, and sex. How shall we treat the measure variable, fixed or random? → fixed effect as we
are interested in the differences between the measures.
Ability = b0 + b1 * measureA + b2 * measureB + …? + …. + e
E ~ N(0, ơe^2)
What shall we do with the child variable, random or fixed? → the children that we
observe here were randomly drawn from a population of children. We are not interested in
these particular children. So using dummy variables for each child and estimating the
differences seems silly. Therefore we use a random effect for the child differences.
Ability = b0 + b1 * measureA + b2 * measureB + childi + …. + eij
Eij ~ N(0, ơe^2)
Childi ~ N(0, ơe^2)
