# pseudo-R squared and multicollinearity checks with beta regression

I am using the package betareg (R software). When I use summary() to know the estimated model, some statistics are displayed. Among these statistics, I have got a pseudo-R squared. However, there is many pseudo R-squareds. Do you know if it is adjusted McFadden's, Efron's, or Cox and Snell's one or an other one?

Moreover, how may I check for multicollinearity using betaregression? Variance inflation factors are not easily calculable, thus I have used first Pearson product-moment correlation coefficients and the condition number (the highest condition index of the correlation matrix). Condition number was 300, above the convention 30 threshold. Thus I standardized two variables that had high condition indices and the condition number was then 5, below the 30 threshold. Should I use standardized variables in my beta regression?

## 1 Answer

The pseudo R-squared is the one suggested by Ferrari & Cribari-Neto for beta regressions: the squared correlation between between the linear predictor for the mean and the link-transformed response. See: calculating a pseudo R2 value when deviance is negative

And for the regressor matrix the usual diagnostics, e.g. based on condition numbers erc., can be applied.

• I should like to sincerely thank you for these helpful comment and link! Your explanation in the link on the R procedure to compute R-squared is simply great! – A. B. Bonache Sep 6 '16 at 7:51