3
$\begingroup$

I'm familiar with multilevel modelling. However, I'm wondering what would happen if you would include level 2 covariates at level 1. If you have, for example, students at level 1 (age, gender,...) and schools (living area, province,...) at level 2 and your dataset looks like this

studentID age gender schoolID living_area province  Y
1         12    1       1        1           1     2.4
2         8     0       1        1           1     3.1
3         10    0       2        0           0     5.2
4         12    0       2        0           0     3.9
5         10    1       3        1           0     4.1
6         9     1       3        1           0     4.8

Why can't I write my model as follows:

$Y_{ij} = \alpha_j + \beta_1*age_{ij} + \beta_2*gender_{ij} + \beta_3*living\_area_{ij} + \beta_4*province_{ij} $ $\alpha_j \sim N(0, \sigma^2_b)$

And why is it better to write my model as follows:

$Y_{ij} = \beta_1*age_{ij} + \beta_2*gender_{ij} + \alpha_j$

$\alpha_j = \alpha_0 + \alpha_1*living\_area_{j} + \alpha_2*province_{j} $

What are the advantages of modelling it that way?

$\endgroup$
1
$\begingroup$

You could go with a mixed formulation, which substitutes the level 2 equation into the level 1 equation:

$Y_{ij} = \alpha_0 + \beta_1*age_{ij} + \beta_2*gender_{ij} + \alpha_1*living\_area_{j} + \alpha_2*province_{j} $

$\alpha_j \sim N(0, \sigma^2_b)$

This correctly signifies that living area and province are measured at the upper level of your data hierarchy (the $_j$ subscript) and shows the common intercept ($\alpha_0$).

Is your question mainly about the model equation or are you also wondering what would happen if you ran a single level OLS regression that included the level 2 predictors in the model?

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.