| OCR Text |
Show 120 heteroskedasticity, but also autocorrelation, which can be a concern particularly with time-series data, although less of a concern in a cross-sectionally dominant data set like mine (King and Roberts 2015). The downside with robust standard errors is that they are a more conservative estimate and thus make it more difficult to obtain statistical significance. In other words, using robust standard errors makes it less likely to commit a Type I error but increases the chances of committing a Type II error. Additionally, I have included both a fixed (FE) and random (RE) effects model for each of the dependent variables. This permits the examination, as well as comparison and contrast, of variation in the electoral fortunes of niche parties both within and between cases.29 In trying to test the strategic interaction model I am continually confronted with the fact that with statistical regression, it tests independent variables while holding all else constant. While this is a challenge, it is nevertheless noticeable that by including independent variables from each cluster, more independent variables are statistically significant across the models than compared to the cluster specific tests of the last section. Additionally, the goodness of fit measures, between, within, and overall Rsquared, are also larger, which indicates that more of the variation in the three dependent variables is being explained by these models and sets of variables. Percent of Vote For this first dependent variable, there are three strategic components that had statistically significant coefficients. First of all, for both the FE and RE models, having coalition experience since the last election does not aid niche parties in the next election. 29 Like for the last series of models, the variance inflation factors (VIFs) were also examined for these models, but the data was not included in the tables. None were found to exceed 10, which is used as a general guideline to indicate when multicollinearity problems might be present. |
