2013
DOI: 10.1080/09645292.2013.855705
|View full text |Cite
|
Sign up to set email alerts
|

Revisiting fixed- and random-effects models: some considerations for policy-relevant education research

Abstract: The use of fixed (FE) and random effects (RE) in two-level hierarchical linear regression is discussed in the context of education research. We compare the robustness of FE models with the modelling flexibility and potential efficiency of those from RE models. We argue that the two should be seen as complementary approaches. We then compare both modelling approaches in our empirical examples. Results suggest a negative effect of special educational needs (SEN) status on educational attainment, with selection i… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

1
71
0

Year Published

2014
2014
2022
2022

Publication Types

Select...
9

Relationship

1
8

Authors

Journals

citations
Cited by 79 publications
(76 citation statements)
references
References 20 publications
(17 reference statements)
1
71
0
Order By: Relevance
“…54 Alternatively, the random-effect estimate itself can be regarded as a precision-weighted estimate. 55 Based on estimates of γ j , we can now proceed to describe hospital performance. For non-linear models, we do this in two different ways.…”
Section: Methodology and Statistical Approachmentioning
confidence: 99%
“…54 Alternatively, the random-effect estimate itself can be regarded as a precision-weighted estimate. 55 Based on estimates of γ j , we can now proceed to describe hospital performance. For non-linear models, we do this in two different ways.…”
Section: Methodology and Statistical Approachmentioning
confidence: 99%
“…Therefore, the equation for the fixed effect (FE) model is as follows: (3) (4) where T and C are time and country specific effects, and μ is the error term. The FE model is useful to investigate the relationship between the predictor and outcome variables within an entity (e.g., country and person), and checks for individual influences on the predictor and outcome variables (Clarke et al 2010).…”
Section: Empirical Methodsmentioning
confidence: 99%
“…Equally, at the higher level, the mean term is no longer constrained by Level 1 effects, so it is free to account for all the higher-level variance associated with that variable. As such, the estimate of b 1 in 12 Instead of using the higher-level unit mean (an aggregate variable), Clarke et al (2010) suggest using global (Diez-Roux 1998) unit characteristics that are correlated with that mean. These global variables express the causal mechanism underlying the association expressed by b 4 , which may not be linear, as is assumed by Models 10 and 11.…”
mentioning
confidence: 99%
“…These global variables express the causal mechanism underlying the association expressed by b 4 , which may not be linear, as is assumed by Models 10 and 11. Including x j would be over-controlling in this case, and such a model has a different interpretation of the higher-level residual, but it is harder to reliably control out all (or even most) of the between effect from the within effect without using x j (Clarke et al 2010) in Equation 11. However, this is not a problem when using the formulation in Equation 12, as the within variable is already group mean centered, so the inclusion of x j is optional depending on the research question at hand, as in the 'within' model in Table 1. 13 Because of this, the number of higher-level units in the sample must be considered, and as such caution should be taken regarding how many higher-level variables (including x j s) the model can estimate reliably.…”
mentioning
confidence: 99%