[Read more...]

Last post on modelling survival data from Veterinary Epidemiologic Research: parametric analyses. The Cox proportional hazards model described in the last post make no assumption about the shape of the baseline hazard, which is an advantage if you have no idea about what that shape might be. With a parametric ... [Read more...]

Next on modelling survival data from Veterinary Epidemiologic Research: semi-parametric analyses. With non-parametric analyses, we could only evaluate the effect one or a small number of variables. To evaluate multiple explanatory variables, we analyze data with a proportional hazards model, the Cox regression. The functional form of the baseline hazard ... [Read more...]

Next topic from Veterinary Epidemiologic Research: chapter 19, modelling survival data. We start with non-parametric analyses where we make no assumptions about either the distribution of survival times or the functional form of the relationship between a predictor and survival. There are 3 non-parametric methods to describe time-to-event data: actuarial life tables, ... [Read more...]

As noted on paragraph 18.4.1 of the book Veterinary Epidemiologic Research, logistic regression is widely used for binary data, with the estimates reported as odds ratios (OR). If it’s appropriate for case-control studies, risk ratios (RR) are preferred for cohort studies as RR provides estimates of probabilities directly. Moreover, it ... [Read more...]

Continuing on the examples from the book Veterinary Epidemiologic Research, we look today at modelling count when the count of zeros may be higher or lower than expected from a Poisson or negative binomial distribution. When there’s an excess of zero counts, you can fit either a zero-inflated model ... [Read more...]

Still going through the book Veterinary Epidemiologic Research and today it’s chapter 18, modelling count and rate data. I’ll have a look at Poisson and negative binomial regressions in R. We use count regression when the outcome we are measuring is a count of number of times an event ... [Read more...]

Next topic on logistic regression: the exact and the conditional logistic regressions. Exact logistic regression When the dataset is very small or severely unbalanced, maximum likelihood estimates of coefficients may be biased. An alternative is to use exact logistic regression, available in R with the elrm package. Its syntax is ... [Read more...]

Third part on logistic regression (first here, second here). Two steps in assessing the fit of the model: first is to determine if the model fits using summary measures of goodness of fit or by assessing the predictive ability of the model; second is to deterime if there’s any ... [Read more...]

Second part on logistic regression (first one here). We used in the previous post a likelihood ratio test to compare a full and null model. The same can be done to compare a full and nested model to test the contribution of any subset of parameters: Interpretation of coefficients Note: ... [Read more...]

We continue to explore the book Veterinary Epidemiologic Research and today we’ll have a look at generalized linear models (GLM), specifically the logistic regression (chapter 16). In veterinary epidemiology, often the outcome is dichotomous (yes/no), representing the presence or absence of disease or mortality. We code 1 for the presence ... [Read more...]

In the previous post, I forgot to show an example of Box-Cox transformation when there’s a lack of normality. The Box-Cox procedure computes values of which best “normalises” the errors. value Transformed value of Y 2 1 0.5 0 -0.5 -1 -2 For example: The plot indicates a log transformation. Matrix Representation We ... [Read more...]

We continue on the linear regression chapter the book Veterinary Epidemiologic Research. Using same data as last post and running example 14.12: Now we can create some plots to assess the major assumptions of linear regression. First, let’s have a look at homoscedasticity, or constant variance of residuals. You can ... [Read more...]

This post will describe linear regression as from the book Veterinary Epidemiologic Research, describing the examples provided with R. Regression analysis is used for modeling the relationship between a single variable Y (the outcome, or dependent variable) measured on a continuous or near-continuous scale and one or more predictor (independent ... [Read more...]

Here are the slides given during the Montreal R User Group workshop on April 30, 2012. RR-MtlRUsers.beamer [Read more...]

A colleague showed me results of his study project with beeswarm plots made by GraphPad. I was wondering if it could be implemented in R and more specifically with ggplot2. There is a R package allowing to draw such graphs, the beeswarm package (beeswa... [Read more...]

Last week a question was asked on the ggplot2 list about using ggplot2 and jags in R (). Here’s what was my answer (a bit updated): Using as an example the school dataset from R2WinBUGS package: Than you can use the mcmcplots package which give a “feel” of ggplot2: ... [Read more...]

Following the post by Arthur Charpentier (Freakonometrics), I wondered what would be the outcome considering my current engagement (veterinary medicine, epidemiology, bovine mastitis). Briefly, Arthur Charpentier’s post looked at clusters of journals publishing the same kind of papers. So I looked at 25 journals (Journal of Dairy Science, Canadian Journal ... [Read more...]