Writing statistical models in equation form Could anybody suggest an appropriate book or perhaps a short online course concerning appropriately writing statistical models in an equation form?  It is so much better to present models this way (i.e., equation form), accompanied by a description, rather than just describing the model in a text form.  
 A: This may not be what you are seeking but the marriage of R and LaTeX gives rise to some excellent opportunities for doing this automatically.  Here is an example:
require(rms)
set.seed(1)
x <- runif(100)
y <- abs(x - 0.5) + runif(100)
f <- ols(y ~ rcs(x, 5))  # restricted cubic spline, 5 default knots
latex(f, file='')

The LaTeX code produced is below, and when you compile into pdf you get the algebraic form of the equation.  This is for a certain class of models and does not allow interactions higher than second order.  The workflow is really elegant when using knitr with R and LaTeX.
\[{\rm E({\rm y}}) = X\beta, {\rm \ \ where} \\ \]
\begin{eqnarray*}
    \lefteqn{X\hat{\beta}=}\\
    & & 0.8507352 \\
    & & + 0.4496578{\rm x}-14.84777({\rm x}-0.08356852)_{+}^{3}+84.03212 ({\rm x}-0.334967)_{+}^{3}  \\
    & &  -109.4476({\rm x}-0.4878107)_{+}^{3}+56.68924 ({\rm x}-0.7303781)_{+}^{3}  \\
    & &   -16.42599({\rm x}-0.9084412)_{+}^{3}  \\
    \end{eqnarray*}
and $ (x)_{+}=x {\rm\ if\ } x>0, \ 0 {\rm\ otherwise} $.

A: Even with a plain old constant-variance regression model, there's a plethora of ways to write it:
$y=\beta_0+\beta_1x_1+\beta_2x_2+...+\beta_px_p+\epsilon$
$E(y)=\beta_0+\beta_1x_1+\beta_2x_2+...+\beta_px_p$
$y=X\beta+\epsilon$
$E(y)=X\beta$
... (and more besides)
and that's not counting adding in the different ways to write down the model for the variance.
A: Maybe a little late to answer this questions, but this short document provides a nice and concise explanation of what each element in an equation for linear and multiple linear regression - with and without interactions - means, and how to move into writing more complex models.
