An analysis based on crashes where there is a harmful event | The (Fatality Analysis Recording System (FARS) data | Why not compare deployment with non-deployment | Not all airbags are equal! | Other factors | The above analyses make an important point | References

Introduction | Data setup and choice of model | Graphical Display | The choice of link function | Modeling the error distribution | Choices required for mixed model fits | Quasibinomial errors | Complementary log-log versus logit link | Details of the model fitting process | Fitted lines, vs fitted normal spline curves | Diagnostic checks | Uniform quantile-quantile plots --- an example. | AIC-based model comparisons | Estimates of $\rho$, and of the dispersion factor | Binomial errors, plus observation level random effects | Confidence intervals for ratios | 99% Lethal time estimates and confidence intervals | What difference does the choice of model make? | Further models and model fitting functions | Binomial errors, observation level random effects, and more | Fits using lme4::glmer() | AIC statistics | Control settings | Gaussian errors, on a complementary log-log scale | Parting comments | The assumed dose-response is a rough approximation! | References

Models for Binary Responses | The Binomial and Poisson --- R functions | More General Models for Binomial-like Counts | Notational conventions | The Pólya urn model as motivation for the betabinomial | Comparing the betabinomial with the quasibinomial | Data that do, and do not, appear binomial | Diseased plants data --- comparison of alternative models | Do differences from the binomial matter? | Numbers of males, in first 12 of 13 children | Bernoulli 0/1 data are a special case | Count data --- Poisson and Related Distributions | Data that do, and do not, appear Poisson | Models for Poisson-Like Counts | References

Plot data | Plot 1988 and 1989 data separately | Correct for control mortality

Introduction | Basic properties of $p$-values | Wear comparison for two shoe materials | What $p$-values do not, and cannot, provide | Strategies for the use of $p$-values | A binary choice is not always appropriate. | One experiment may not, on its own, be enough | Small differences may be of no interest | Effectiveness of soporofic drugs | A test that sets $\mu$ = 0.8 hours as the baseline | Likelihood ratio and false positive risk | What is the definitive question? | Density curves, under the NULL, and under the alternative | Maximum likelihood ratio versus $p$-value | False positive risk versus $p$-value | Power -- how well does a planned experiment discriminate? | The power of a $t$ or other statistical test | False positive risk, when $\alpha$ is used as cutoff | What choices of cutoff $\alpha$, and of power, make sense? | How should results be reported? | Further reading and references | References

Handling skew distributions and/or outliers | Use tree-based methods appropriately! | Choosing the split point | Random forest fit to residuals from a trend surface | Linear and quadratic discriminant analysis | Tree-based methods | Inclusion of interaction terms, with lda() | Multiple Levels of variation -- the Vowel dataset

Prepare data -- Weekly and daily counts | Fits, to the daily data, and to the weekly data

References