Multilevel Workshop

“Listening to the world. Well, I did that, and I still do it. I still do it.” (Mary Oliver in Oliver and Tippett 2015)

1 Introduction

This site contains materials for a workshop on multilevel modeling.

1.1 Background

Multilevel models are useful when you have data that are nested or clustered inside social units such as schools, neighborhoods, states, or countries.

Multilevel models are also useful when you have longitudinal data where repeated measures are collected for study participants.

The Importance of Multilevel Models

Multilevel models may improve one’s statistical inferences in two important substantive ways.

• Multilevel models adjust standard errors for clustering, and thus calculate appropriate p values. Failure to use a model that accounts for the clustering in the data may lead to improperly calculated standard errors and p values, possibly leading to false attributions of statistical significance (false positives) (see Chapter 2).
• Multilevel models adjust regression coefficients (\(\beta\)’s) for the presence of clustering. Failure to use a model that accounts for the clustering in the data may lead to improperly calculated regression coefficients (\(\beta\)’s) which may have the wrong magnitude, the wrong statistical significance, and even the wrong sign (see Chapter 3).

1.2 Simulated Multilevel Data

The data used in these workshop materials are simulated data on parents, children and families. The data are simulated to come from 30 hypothetical countries around the world. These are the same data used and discussed in my book Multilevel Thinking: Discovering Variation, Universals, and Particulars in Cross-Cultural Research.

There are two versions of the data: a cross-sectional data set from a single point in time; a longitudinal version of the data spanning several time points.

The Data Can Be Downloaded Here:

• Cross Sectional Data
• Longitudinal Data

Figure 1.1: Countries of the World

Table 1.1: Variables in Simulated Multilevel Data

pos	variable	label
1	country	country id
2	HDI	Human Development Index
3	family	family id
4	id	unique country family id
5	identity	hypothetical identity group variable
6	intervention	recieved intervention
7	physical_punishment	physical punishment in past week
8	warmth	parental warmth in past week
9	outcome	beneficial outcome

Table 1.2: Sample Data From Simulated Multilevel Data

country	HDI	family	id	identity	intervention	physical_punishment	warmth	outcome
21	84	98	21.98	1	0	4	1	52.38
23	73	95	23.95	0	1	2	5	47.17
12	33	39	12.39	0	0	3	4	41
10	59	68	10.68	1	0	2	3	51.36
7	73	74	7.74	1	1	0	5	65.49

Oliver, Mary, and Krista Tippett. 2015. “Mary Oliver: ‘I Got Saved by the Beauty of the World.’” The On Being Project. https://onbeing.org/programs/mary-oliver-i-got-saved-by-the-beauty-of-the-world/.