Assess how much of the error in prediction is due to lack of model fit.
Value
ols_pure_error_anova returns an object of class
"ols_pure_error_anova". An object of class "ols_pure_error_anova" is a
list containing the following components:
- lackoffit
lack of fit sum of squares
- pure_error
pure error sum of squares
- rss
regression sum of squares
- ess
error sum of squares
- total
total sum of squares
- rms
regression mean square
- ems
error mean square
- lms
lack of fit mean square
- pms
pure error mean square
- rf
f statistic
- lf
lack of fit f statistic
- pr
p-value of f statistic
- pl
p-value pf lack of fit f statistic
- mpred
data.framecontaining data for the response and predictor of themodel- df_rss
regression sum of squares degrees of freedom
- df_ess
error sum of squares degrees of freedom
- df_lof
lack of fit degrees of freedom
- df_error
pure error degrees of freedom
- final
data.frame; contains computed values used for the lack of fit f test
- resp
character vector; name of
response variable- preds
character vector; name of
predictor variable
Details
The residual sum of squares resulting from a regression can be decomposed into 2 components:
Due to lack of fit
Due to random variation
If most of the error is due to lack of fit and not just random error, the model should be discarded and a new model must be built.
Note
The lack of fit F test works only with simple linear regression. Moreover, it is important that the data contains repeat observations i.e. replicates for at least one of the values of the predictor x. This test generally only applies to datasets with plenty of replicates.
The test is not available for weighted models, because the regression sum of squares would come from the weighted model while the lack of fit and pure error sums of squares are computed from unweighted residuals.
A fixed offset (an offset() term in the formula or the offset argument of
lm()) is known rather than estimated, so it is removed from the response
before the sums of squares are split. The decomposition then applies to the
offset-adjusted response and the error sum of squares equals
deviance(model); mpred holds the adjusted response.
References
Kutner, MH, Nachtscheim CJ, Neter J and Li W., 2004, Applied Linear Statistical Models (5th edition). Chicago, IL., McGraw Hill/Irwin.
Examples
model <- lm(mpg ~ disp, data = mtcars)
ols_pure_error_anova(model)
#> Lack of Fit F Test
#> -----------------
#> Response : mpg
#> Predictor: disp
#>
#> Analysis of Variance Table
#> ----------------------------------------------------------------------
#> DF Sum Sq Mean Sq F Value Pr(>F)
#> ----------------------------------------------------------------------
#> disp 1 808.8885 808.8885 76.51266 9.380327e-10
#> Residual 30 317.1587 10.57196
#> Lack of fit 25 304.2787 12.17115 4.724824 0.04563623
#> Pure Error 5 12.88 2.576
#> ----------------------------------------------------------------------