The function print.summary.lm
produces a typical table of coefficients, standard errors and
p-values along with “significance stars”. In addition, a table of overall p-values are shown.
Multi-Parameter Regression (MPR) models are defined by allowing mutliple distributional parameters to
depend on covariates. The regression components are:
$$g_1(\lambda) = x^T \beta$$
$$g_2(\gamma) = z^T \alpha$$
$$g_3(\rho) = w^T \tau$$
and the table of coefficients displayed by print.summary.lm
follows this ordering.
Furthermore, the names of the coefficients in the table are proceeded by “.b
” for
\(\beta\) coefficients, “.a
” for \(\alpha\) coefficients and “.t
” for
\(\tau\) coefficients to avoid ambiguity.
Let us assume that a covariate \(c\), say, appears in both the \(\lambda\) and \(\gamma\)
regression components. The standard table of coefficients provides p-values corresponding to the following
null hypotheses:
$$H_0: \beta_c = 0$$
$$H_0: \alpha_c = 0$$
where \(\beta_c\) and \(\alpha_c\) are the regression coefficients of \(c\) (one for each of the
two components in which \(c\) appears). However, in the context of MPR models, it may be of interest
to test the hypothesis that the overall effect of \(c\) is zero, i.e., that its \(\beta\)
and \(\alpha\) effects are jointly zero:
$$H_0: \beta_c = \alpha_c = 0$$
Thus, if overall=TRUE
, print.summary.lm
displays a table of such “overall p-values”.