Deviance (statistics)

In statistics, deviance is a goodness-of-fit statistic for a statistical model; it is often used for statistical hypothesis testing. It is a generalization of the idea of using the sum of squares of residuals in ordinary least squares to cases where model-fitting is achieved by maximum likelihood. It plays an important role in exponential dispersion models and generalized linear models. In statistics, deviance is a goodness-of-fit statistic for a statistical model; it is often used for statistical hypothesis testing. It is a generalization of the idea of using the sum of squares of residuals in ordinary least squares to cases where model-fitting is achieved by maximum likelihood. It plays an important role in exponential dispersion models and generalized linear models. The unit deviance d ( y , μ ) {displaystyle d(y,mu )} is a bivariate function that satisfies the following conditions: The total deviance D ( y , μ ^ ) {displaystyle D(mathbf {y} ,{hat {oldsymbol {mu }}})} of a model with predictions μ ^ {displaystyle {hat {oldsymbol {mu }}}} of the observation y {displaystyle mathbf {y} } is the sum of its unit deviances: D ( y , μ ^ ) = ∑ i d ( y i , μ ^ i ) {displaystyle D(mathbf {y} ,{hat {oldsymbol {mu }}})=sum _{i}d(y_{i},{hat {mu }}_{i})} . The (total) deviance for a model M0 with estimates μ ^ = E [ Y | θ ^ 0 ] {displaystyle {hat {mu }}=E} , based on a dataset y, may be constructed by its likelihood as: Here θ ^ 0 {displaystyle {hat { heta }}_{0}} denotes the fitted values of the parameters in the model M0, while θ ^ s {displaystyle {hat { heta }}_{s}} denotes the fitted parameters for the saturated model: both sets of fitted values are implicitly functions of the observations y. Here, the saturated model is a model with a parameter for every observation so that the data are fitted exactly. This expression is simply 2 times the log-likelihood ratio of the full model compared to the reduced model. The deviance is used to compare two models – in particular in the case of generalized linear models (GLM) where it has a similar role to residual variance from ANOVA in linear models (RSS). Suppose in the framework of the GLM, we have two nested models, M1 and M2. In particular, suppose that M1 contains the parameters in M2, and k additional parameters. Then, under the null hypothesis that M2 is the true model, the difference between the deviances for the two models follows an approximate chi-squared distribution with k-degrees of freedom. Some usage of the term 'deviance' can be confusing. According to Collett: Hypothesis testing on the deviance can use Wilks' theorem. The unit deviance for the Poisson distribution is d ( y , μ ) = 2 ( y log ⁡ y μ − y + μ ) {displaystyle d(y,mu )=2left(ylog {frac {y}{mu }}-y+mu ight)} , the unit deviance for the Normal distribution is given by d ( y , μ ) = ( y − μ ) 2 {displaystyle d(y,mu )=left(y-mu ight)^{2}} .