16.12.2017 Views

Fall 2017 JPI

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

While MSG do add some value with their analysis of the effect of ethnic cleavages, there are<br />

some significant flaws in their models that limit the usefulness of their results. Mainly, in Models 3<br />

and 4 where MSG include interactive terms, they fail to include all of the terms denoting all possible<br />

relationships between the independent variables. According to MSG, the fully-specified model is as<br />

follows:<br />

Parties = δ0 + δ1 Fragmentation<br />

+δ2Concentration<br />

+δ3log(magnitude)<br />

+δ4 Fragmentation × Concentration ×log(Magnitude)<br />

+δ8 Proximity × PresidentialCandidates + ε<br />

But, as GBC point out, the actual accurate fully specified model should be the following:<br />

Parties = β0 + β1 Fragmentation<br />

+β2Concentration<br />

+β3log(Magnitude)<br />

+β4 Fragmentation × Concentration<br />

+β5 ∗ Fragmentation × log(Magnitude)<br />

+β6 ∗ Concentration × log(Magnitude)<br />

+β7 Fragmentation × Concentration ×log(Magnitude)<br />

+β8 Proximity<br />

+β9 PresidentialCandidates<br />

+β10 Proximity × PresidentialCandidates + ε<br />

In MSG’s model they omit Fragmentation X log(Magnitude), Fragmentation X Concentration,<br />

Concentration X log(Magnitude), and Presidential Candidates. This is a key error in that when using interactive<br />

models, it is essential to include every possible constitutive term. 33 By omitting those four terms, MSG<br />

are assuming that those coefficients are approximately zero but there is not any substantial evidence<br />

that that is true. Furthermore, in the instance that any of those coefficients are not truly zero, that<br />

would cause all of the estimators to be biased and force the inaccurate model to pass through the<br />

origin regardless of the data’s true tendencies. Thus, in Table 3 below, I present the results side by<br />

side of each model (institutional, sociological, additive, and interactive) using MSG’s original model<br />

on the left, and Golder’s corrected models on the right. Note that a corrected version of the purely<br />

sociological model is not included because Golder simply corrected some inconsistencies in MSG’s<br />

33 Brambor, Clark, and Golder, African party systems, 2006.<br />

<strong>JPI</strong> <strong>Fall</strong> <strong>2017</strong>, pg. 61

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!