Readability improvement

docs/src/examples.md

@@ -42,6 +42,49 @@ julia> round.(predict(ols), digits=5)
  1.83333
  4.33333
  6.83333
+
+julia> round.(confint(ols); digits=5)
+2×2 Matrix{Float64}:
+ -8.59038  7.25704
+ -1.16797  6.16797
+
+julia> round(r2(ols); digits=5)
+0.98684
+
+julia> round(adjr2(ols); digits=5)
+0.97368
+
+julia> round(deviance(ols); digits=5)
+0.16667
+
+julia> dof(ols)
+3
+
+julia> dof_residual(ols)
+1.0
+
+julia> round(aic(ols); digits=5)
+5.84252
+
+julia> round(aicc(ols); digits=5)
+-18.15748
+
+julia> round(bic(ols); digits=5)
+3.13835
+
+julia> round(dispersion(ols.model); digits=5)
+0.40825
+
+julia> round(loglikelihood(ols); digits=5)
+0.07874
+
+julia> round(nullloglikelihood(ols); digits=5)
+-6.41736
+
+julia> round.(vcov(ols); digits=5)
+2×2 Matrix{Float64}:
+  0.38889  -0.16667
+ -0.16667   0.08333
 ```
 
 ## Probit regression

docs/src/index.md

@@ -35,7 +35,7 @@ functions are
             Binomial (LogitLink)
                Gamma (InverseLink)
      InverseGaussian (InverseSquareLink)
-    NegativeBinomial (LogLink)
+    NegativeBinomial (NegativeBinomialLink, often used with LogLink)
               Normal (IdentityLink)
              Poisson (LogLink)
 
@@ -147,20 +147,74 @@ F-test: 2 models fitted on 50 observations
 ## Methods applied to fitted models
 
 Many of the methods provided by this package have names similar to those in [R](http://www.r-project.org).
-- `coef`: extract the estimates of the coefficients in the model
+- `adjr2`: adjusted R² for a linear model (an alias for `adjr²`)
+- `aic`: Akaike's Information Criterion
+- `aicc`: corrected Akaike's Information Criterion for small sample sizes
+- `bic`: Bayesian Information Criterion
+- `coef`: estimates of the coefficients in the model
+- `confint`: confidence intervals for coefficients
+- `cooksdistance`: [Cook's distance](https://en.wikipedia.org/wiki/Cook%27s_distance) for each observation, giving an estimate of the influence of each data point. Currently only implemented for linear models without weights.
 - `deviance`: measure of the model fit, weighted residual sum of squares for lm's
+- `dispersion`: dispersion (or scale) parameter for a model's distribution
+- `dof`: the number of degrees of freedom consumed in the model
 - `dof_residual`: degrees of freedom for residuals, when meaningful
+- `fitted`: the fitted values of the model
 - `glm`: fit a generalized linear model (an alias for `fit(GeneralizedLinearModel, ...)`)
 - `lm`: fit a linear model (an alias for `fit(LinearModel, ...)`)
-- `r2`: R² of a linear model or pseudo-R² of a generalized linear model
+- `loglikelihood`: log-likelihood of the model
+- `modelmatrix`: design matrix
+- `nobs`: number of rows, or sum of the weights when prior weights are specified
+- `nulldeviance`: deviance of the linear model which includes the intercept only
+- `nullloglikelihood`: log-likelihood of the linear model which includes the intercept only
+- `predict`: obtain predicted values of the dependent variable from the fitted model
+- `r2`: R² of a linear model (an alias for `r²`)
+- `residuals`: vector of residuals from the fitted model
+- `response`: model response (a.k.a the dependent variable)
 - `stderror`: standard errors of the coefficients
-- `vcov`: estimated variance-covariance matrix of the coefficient estimates
-- `predict` : obtain predicted values of the dependent variable from the fitted model
-- `residuals`: get the vector of residuals from the fitted model
+- `vcov`: variance-covariance matrix of the coefficient estimates
+
 
 Note that the canonical link for negative binomial regression is `NegativeBinomialLink`, but
 in practice one typically uses `LogLink`.
 
+```jldoctest methods
+julia> using GLM, DataFrames;
+
+julia> data = DataFrame(X=[1,2,3], y=[2,4,7]);
+
+julia> test_data = DataFrame(X=[4]);
+
+julia> mdl = lm(@formula(y ~  X), data);
+
+julia> round.(coef(mdl); digits=8)
+2-element Vector{Float64}:
+ -0.66666667
+  2.5
+
+julia> round(r2(mdl); digits=8)
+0.98684211
+
+julia> round(aic(mdl); digits=8)
+5.84251593
+```
+`predict` method returns predicted values of response variable from covariate values `newX`.
+If you ommit `newX` then it return fitted response values. You will find more about [predict](https://juliastats.org/GLM.jl/stable/api/#StatsBase.predict) in the API docuemnt.
+
+```jldoctest methods
+julia> round.(predict(mdl, test_data); digits=8)
+1-element Vector{Float64}:
+ 9.33333333
+```
+`cooksdistance` method computes [Cook's distance](https://en.wikipedia.org/wiki/Cook%27s_distance) for each observation in linear model `obj`, giving an estimate of the influence of each data point. Currently only implemented for linear models without weights.
+
+```jldoctest methods
+julia> round.(cooksdistance(mdl); digits=8)
+3-element Vector{Float64}:
+ 2.5
+ 0.25
+ 2.5
+```
+
 ## Separation of response object and predictor object
 
 The general approach in this code is to separate functionality related

src/GLM.jl

@@ -12,14 +12,15 @@ module GLM
     import Statistics: cor
     import StatsBase: coef, coeftable, confint, deviance, nulldeviance, dof, dof_residual,
                       loglikelihood, nullloglikelihood, nobs, stderror, vcov, residuals, predict,
-                      fitted, fit, model_response, response, modelmatrix, r2, r², adjr2, adjr², PValue
+                      fitted, fit, model_response, response, modelmatrix, r2, r², adjr2, adjr², PValue,
+                      aic, aicc, bic
     import StatsFuns: xlogy
     import SpecialFunctions: erfc, erfcinv, digamma, trigamma
     import StatsModels: hasintercept
     export coef, coeftable, confint, deviance, nulldeviance, dof, dof_residual,
            loglikelihood, nullloglikelihood, nobs, stderror, vcov, residuals, predict,
            fitted, fit, fit!, model_response, response, modelmatrix, r2, r², adjr2, adjr²,
-           cooksdistance, hasintercept
+           cooksdistance, hasintercept, aic, aicc, bic, dispersion
 
     export
         # types