12

As with many other cases in statistics, the goal of finding a single test to replace one's judgement is a bad one. There are several sources of information you can and should use while deciding: the theoretical expectation of the distribution, prior empirical work on the topic, the properties of the data (e.g. is it truncated or zero-inflated?), and the ...


10

Another great answer from @mkt on this forum. Here are a few more pointers you might find useful. GLMs include some widely used types of regression models: Binary Logistic Regression Models; Binomial Logistic Regression Models; Multinomial Logistic Regression Models; Ordinal Logistic Regression Models; Poisson Regression Models; Beta Regression Models; ...


5

I would generally recommend your option #4, collapsing the two factors to a one-way layout that includes all of the possible combinations. That way you can fit models without (1) throwing away cases with "missing" data or (2) running into rank-deficiency problems. You can still estimate all the things you want (e.g. difference between W and WO for ...


5

1) You do not need the raw data to be normally distributed. It's only the residuals that need to be. 2) Removing 'outliers' is generally a bad idea unless you have very good reason to believe that those data points are invalid for some reason, such as instrument failure. 3) If the residual distribution is actually a problem, you can still avoid log ...


4

This was a reply to @Victor's comment on @mkt's answer, but it grew rather large, and I suppose it answers the question. The point of using a GLM is to allow different error distributions than Gaussian. Is the data generating process continuous, with a central tendency and can it take on both positive and negative values? Then a regular LM is a decent ...


4

In the following linked paper you have a detailed answer: in short, MLE will try to set the estimated coefficients to +/inf, as explained at pag.340-341 of this paper that is a very good suggested reading. If you need practical remedies, and examples/solutions to be implemented in R consider also this answer in addition to the previous text. And this ...


3

As Isabella Ghement also mentioned in her comments, it seems that you have a count variable. Hence, you could instead try fitting a Poisson or negative binomial mixed effects model. Both typically specify a log link function, resulting in the same type of model for the mean as the one you specify with the Gamma mixed model. In some implementations, such as ...


3

A famous adage states that sample size is where you randomize. It seems that you have randomly assigned (?) 4 plants to treatment and 5 plants to control. So even if you go for the suggested Wilcoxon's rank sum test on the differences, you would have a fairly low overall sample size of 9 plants. Additionally, it it usually the case that nonparametric ...


2

Yes, it is true. And, I can't think of an exception when that would not be the case. But, you have to watch out about your data set, and when you test your model against new data. Let's say variable A has a standardized coefficient of 1.0 and variable B has one of 0.75. So, variable A is more influential than variable B. However, in your new data set ...


1

You ask: Would you rely on the log-transformation? Or would you rather use a non-parametric method? for me, that would depend on what the variables are and what my question is. Does taking the log make substantive sense? It often does make sense for variables involving money (such as income, wealth, expenditures) because we tend to think of those ...


1

Stepwise variable selection tends to provide too optimistic results (too low p values etc). The main critique with the method is that researchers often ignore that fact and present the model results without mentioning that bias. In your comparison, the focus is not on how valid the results are but on how the method competes with alternative modelling ...


1

According to the Frequentists' theory and MLE, the model and other following statistical tests only work correctly when it follows the real underlying data generating distribution. If the data is sampled by Poisson distribution from your project context, Poison regression should be used rather then Ordinal regression. Firstly, Poisson distribution is the ...


1

A couple of points: It seems to me that you could your outcome as a count, i.e., the number of frog callings in a three hours period. Hence, you could fit a mixed effects Poisson or negative binomial models (the latter accounts for over-dispersion). These are, for example, provided by the GLMMadaptive package I have written; you can find several examples ...


1

The "base" level (often called the 'reference level') should not have been left out of the model. It is represented by the intercept. The other levels are typically specified in your output, but those coefficients are not actually the values for those levels, instead they are the differences between the values for the indicated level and the base level. ...


1

I() means "as is" (see ?I): this is needed because operators such as ^ and * have different meanings in a formula context than when they are doing regular computation. I(age^2) and I(age*age) are equivalent; they both mean to add an "age-squared" term to the model. If the formula contained unprotected age^2 or age*age terms, it would denote the interaction ...


1

The relative importance is based on the coefficients, the scaled importance is the relative importance scaled between 0 and 1. You can see the normalized coefficients for comparison by using h2o.coef_norm().


1

For the record, a simple pure R implementation of R's glm algorithm, based on Fisher scoring (iteratively reweighted least squares), as explained in the other answer is given by: glm_irls = function(X, y, weights=rep(1,nrow(X)), family=poisson(log), maxit=25, tol=1e-16) { if (!is(family, "family")) family = family() variance = family$variance ...


Only top voted, non community-wiki answers of a minimum length are eligible