provide tutorial material for HLM

[hannesdatta]

provide by @mpachali to @srosh2000

[mpachali]

• Hi @srosh2000 , some input:

• Bayesian version of the model (also with application using simulated data):

https://rdrr.io/cran/bayesm/man/rhierLinearMixture.html

• Detailed explanations in Bayesian Statistics and Marketing (Rossi et al. 2005):

See Section 3.7, page 70 for extensive model description and inference. Please let me know if you need access.

• The LME4 package in R (see http://www.alexanderdemos.org/Class13.html for some discussion) seems to implement HLM using traditional maximum likelihood methods. Might be easier at the start, but let's see...

• The idea of the HLM is to relate regression equations (over consumers) through correlations between the regression coefficients. An alternative is to relate regressions through correlated errors (SUR-models).

• This might be a good model for us as well, but I guess HLM are more suitable if the focus is getting i-level estimates

• A good paper for how to implement SUR-models with larger N is:

Lidan Tan, Khai Xiang Chiong & Hyungsik Roger Moon (2021) Estimation of high-dimensional seemingly unrelated regression models, Econometric Reviews, 40:9, 830-851, DOI: 10.1080/07474938.2021.1889195

• Please let me if you have questions

[srosh2000]

@mpachali Thanks for the inputs, was useful going through them. I'd like to still clarify the exact model specification in our context?

• DV: no. of books added per active user for starters ( and there are 3 other as per the thesis)
• IV: OxCGRT stringency index per country For the random intercept/ slope models: What are our levels? Level 1 is users and level 2 is country? so the random effect variable would be country?