Ignoring multilevel data structure in favor of single-level models can lead to misleading conclusions.
If we analyze the multilevel data with a single level regression model, both the estimate of a slope and the estimate of a standard error of slope are biased.
Case:
A study was conducted to find a relationship between students’ socioeconomic status and students’ academic achievement.
Data:
– Multilevel data with 500 students from 50 high schools.
– 10 students were sampled from each high school.
Model:
– Simple regression model
i indicates students (i = 1, 2, …10)
j indicates schools (j = 1, 2, …50)
Results:
– the estimate of slope is biased.
– the estimate of standard error of the slope is underestimated.
