NEWS.md
show.bootstraps defaults to TRUE only if fewer than 100,000 observationsglm objectsinzplot (from iNZightPlots)exponentiate.cis argument to iNZightSummary which replaces the CIs with exponentiated versions if appropriate (FALSE by default)Release date: 21 April 2020
stringsAsFactors = TRUE for upcoming R 4.0.0exluded variablesRelease date: 11 November 2019
Release date: 2 September 2019
factorComp() function to obtain adjusted pairwise comparisons of factor levels in a modelRelease date: 15 July 2019
y ~ 1)Release date: 30 April 2019
loess() callsRelease date: 01 February 2019
Release date: 02 October 2017
Release date: 23 August 2017
This isn’t a hugely updated version, however fixing up a bunch of bugs to make the Model Fitting module better (over on iNZightModules).
subset argument to lm (via update()) to perform bootstraps, rather than the long-winded data-bootstrapping call-modifying version that was buggyRelease date: 9 January 2017
glm objectRelease date: 20 July 2015
Release date: 27 March 2014
Residual summary plots from plotlm6 can now make use of the iNZightPlots graphics rather than the defaults. It requires the user to have iNZightPlots installed to work, but reverts to the old plots if it is not.
In the new grid based plots, quantile smoothers are used rather than loess smoothers. This greatly increases efficiency when large data sets are analysed.
Release date: 18 January 2014
Support for generalized linear models (GLMs) and svyglm objects from the survey package.
Changes to the iNZightSummary output include:
Output now hides output of confounding variables through the exclude argument, and lists these at the top of the output.
Displays the type of fit (e.g., Survey / Generalized Linear / Model).
The QQ-plot array has been replaced by a single plot with the parametric bootstrap data QQ-plots all on a single plot, overlaid by the QQ-plot of the true data (this was suggested by Thomas Lumley).
grid, and minimizes margin whitespace and draws simulated histograms in a different colour.The bootstrap models functions have been re-written to account for the design option in survey GLMs, as well as the case when the GLM binomial response is SUCCESS / N.TRIALS. This caused errors in the fit$model that was previously being used.
The bootstrap lines from plotlm6 have been fixed so that they now work for (svy)glm objects. There is also an optional cut-off if the sample size becomes too small (which can be overridden by the showBootstraps = TRUE|FALSE argument.
factorMeans and adjustedMeans have been enhanced to work with GLMs.
survey package’s svyglm.First release of new package. Contains model fitting subset used for the iNZight package.
Added histogramArray and qqplotArray plots to show how residuals from a model compare to the residuals generated from that model.
New margin of error calculation functions. Initially written by Danny Chang. Used for comparison between levels of a factor. moecalc has a few standard methods that can be used: print, plot, and summary. In addition, a multicomp method has been added which is a useful tabulation of multiple comparison output. Note however with multicomp that the p-values are currently unadjusted.
New summary output, iNZightSummary. Includes several Changes compared to the R-base model summary output. These include the following:
Now showing the factor itself in the output, not just rows for coefficients for levels of the factor.
When a factor is included in a model, the summary output will show the name of the factor and show the p-value for the factor (based on Type-III sums of squares). This p-value is not affected by further use of the factor (i.e. in an interaction). Sometimes this p-value cannot be calculated (i.e. when there are unobserved factor level combinations) and the p-value will be omitted.
The baseline level of a factor is now shown, with an estimate of 0.
All p-value output for levels of a factor is indented to the right by two characters to distinguish it from being a level.
The output for each factor level is now just the level name and not the name of the variable concatenated with the level name. The level name is also indented by two characters, again to distinguish it from the variable itself.
Removing F-statistic and associated p-value as it’s mostly useless. It only shows us whether nothing is correlated with the response variable, i.e. whether we’re completely wasting our time.
Removed model call output and residual output. Replaced with “Model for:” (plus response name).
Added a new plot.lm function. The main difference being that it includes bootstrapped smoothers in its output as well as the regular trend lines. Also includes plots based on the s20x package’s normcheck function.
Added partial residual plots. Most useful for determining whether the inclusion of a transformation of a variable is necessary. For example, adding a logged or polynomial explanatory variable to the model.
Adding bootstrapped estimates to iNZightSummary. Accessed by calling the function with method="bootstrap".