= Log-linear model
{title2=$\log\mu=X\beta$}
A <log-linear model> makes the <logarithm> of a positive <expectation> linear in model <parameters>. For <independent> cell counts with a <Poisson distribution> $N_a$ in a <contingency table>, write $\mu_a=\mathbb EN_a$ and $\log\mu_a=x_a^T\beta$. This is a <Poisson regression> with <log link>; finite linear predictors give strictly positive fitted cell <expectations>.
For a two-way table, the additive model
$$
\log\mu_{ij}=\alpha+r_i+c_j
$$
implies $\mu_{ij}=e^\alpha e^{r_i}e^{c_j}$. Conditional on the total count, the cell counts have a <multinomial distribution> whose cell <probabilities> factor into row and column <probabilities>; this is <independence> of the two categorical classifications. For example, the cell-mean table $\left(\begin{smallmatrix}2&4\\3&6\end{smallmatrix}\right)$ has common column ratio two and <odds ratio> $(2\cdot6)/(4\cdot3)=1$.
Adding interaction terms $\gamma_{ij}$ to the linear predictor permits association. In a $2\times2$ table the association is measured by
$$
\log\frac{\mu_{11}\mu_{22}}{\mu_{12}\mu_{21}}
=\log\mu_{11}+\log\mu_{22}-\log\mu_{12}-\log\mu_{21}.
$$
It vanishes in the additive model. Hence additive main effects describe one particular <log-linear model>; <log-linear models> also include interactions and more general <design matrices>. <Independent> Poisson sampling and multinomial sampling conditional on the total use different sampling schemes, while producing this same <independence> structure for the normalized cell <probabilities>.
Back to article page