Multi-level modeling with mixed command

[Vincent KH Lee]

I'm trying to run a mixed command with a hierarchical panel data.

There are 3 levels total,

Lv 3. Country , Lv 2. Brand, Lv 1. Individual subject

From previous literature, I discovered that when using the panel data, the first level should be time-series component.


Accordingly, that would mean I need 4 levels total,

Lv 4. Country, Lv 3. Brand, Lv 2. Individual subject, Lv 1. Time


The main goal here is to estimate the impact of certain event (time) on the time-invariant covariate in Lv 2. individual subject.


In this case, which out of the three stata commands best describe the specified model?

# 1.

mixed y i.Treatment##i.Lv2-Covariate, || Country:, || Brand:, cov(unstructured)

# 2.

mixed y i.Treatment##i.Lv2-Covariate, || Country:, || Brand:, || Individual: Lv2-Covariate, cov(unstructured)

# 3.

mixed y i.Treatment##i.Lv2-Covariate, || Country:, || Brand:, || Individual: Lv2-Covariate, || time:, cov(unstructured)

I'm pretty new to multi-level model with random effects, but as per my understanding goes, commands #2 and #3 involves cross-level interactions.

To summarize, should I include the time & individual level in the command as in #2 and #3? given the dataset is 2-years worth of strongly balanced panel?

or specifying it 'till the Brand level is enough?

Any help or advice is appreciated in advance.

Thank you.