11,111 reputation
13163
bio website fromthebottomoftheheap.net
location Regina, Canada
age 38
visits member for 4 years, 2 months
seen 2 hours ago

I'm Quantitative Environmental Scientist in the Institute of Environmental Change & Society, at the University of Regina, Canada. I undertake research on environmental problems, including climate change and atmospheric pollution, affecting lakes. I use lake sediments to look back in time at the history of lakes to look at what organisms are present and how the species in the lake have changed through time and how lakes evolve and respond to pollution and perturbations.

I'm also an Adjunct Professor in the Department of Biology at the University of Regina.


2h
comment How to implement reduced-rank regression in R?
@amoeba we may be talking about slightly different methods - RDA gets called a lot of things. We implement it in rda() via QR decomposition and SVD for efficiency, but that method gets the same result as the R code I showed in the comment earlier. Which makes me think what we do, which has been called reduced rank regression, is not the reduced rank regression the OP is looking for :-)
5h
comment How to implement reduced-rank regression in R?
I'd be surprised is VGAM didn't do this; it has plenty of continuous distribution family functions (though note I haven't looked in detail at the RRR function in VGAM recently). You can also do something that is known as reduced rank regression with the vegan package. We call this Redundancy Analysis (RDA) but it also goes by the name reduced rank regression. And as @amoeba says, RDA can be computed by doing fit <- fitted(lm(Y ~ X, data = foo)) then prcomp(fit). If this is what you want, then rda() in vegan would be a good start.
Nov
18
comment impose an intercept on lm in r
You can add offset in lm() too. Was there a reason to switch to glm() here?
Nov
18
comment impose an intercept on lm in r
Does it matter if the intercept moves around a bit? I presume that the estimate of this constant term is so uncertain as to have a confidence interval on its estimate that includes 0? Just because physics dictates that the intercept be a certain value, doesn't mean you have to force exactly that value. By forcing that value exactly, you may induce bias in the estimation of other parameters of the model. The point is that you have noise in your measurements and you can account for that or at least look at the effect of that on the estimates of the constant term.
Nov
17
comment Can mean plus one standard deviation exceed maximum value?
Why do you want to add (or subtract) one standard deviation from the mean? The SD is a measure of the spread of the data. Did you want the standard error of the mean instead perhaps?
Nov
10
comment weighted disease prevalence in logistic GAM
@John could you use an offset() term to include the population? That's what I might do with a Poisson model for example if I wanted to standardise for effort or sample area etc.
Nov
10
answered weighted disease prevalence in logistic GAM
Nov
7
comment Calculate PCoA scores for dataframe “x”, based on the distance matrix of dataframe “y”
I think I understand conceptually what you want to do, at a higher level, but you can't really do anything to a dataset with the dissimilarities from the first data set. I explain a directly analogue of what you want to do using a constrained PCoA in my answer and suggest an alternative which avoids even doing the embedding in principal coordinates.
Nov
7
answered Calculate PCoA scores for dataframe “x”, based on the distance matrix of dataframe “y”
Nov
4
comment backward selection but regression coefficients not significative?
Depends what you by "in R"? The default step() function doesn't explicitly use the $p$ value at all. Instead it uses AIC to do selection. That may well explain what the OP is seeing, as may other things. We need more details to answer this.
Nov
4
comment backward selection but regression coefficients not significative?
How were you doing the backwards elimination? What functions, what test statistic or selection method, AIC etc? How did you generate the table of "results" shown? These are important details.
Nov
4
awarded  Notable Question
Nov
3
comment How to measure goodness of fit in a simple quadratic Gaussian GLM?
In terms of the figure you show, a linear fit would do just as well as the quadratic. Then you could compare the two fits with AIC or a likelihood ratio. Perhaps you could explain in more detail what the real problem is that you wish to solve. Also, why glm() when this is just a linear model? It is far more efficient to fit via lm().
Nov
3
comment How to measure goodness of fit in a simple quadratic Gaussian GLM?
Have you plotted the data & the model fit? Have you plotted residuals and other model diagnostics? I can easily foresee situations where you may have a good model but a lowish $R^2$-alike.
Nov
3
comment Generalized additive models — who does research on them besides Simon Wood?
@Flounderer I don't think Gung was suggesting that such questions weren't useful, just not on topic for Cross Validated. This is a general statement for the family of Stack Exchange sites, not something we dreamt up just for Cross Validated. This isn't a discussion site and there really isn't going to be a definitive answer etc. I'm sympathetic as I think this is an interesting question too, but I also know the Stack Exchange rules; that's why I threw in a bone in the comments in case the OP didn't focus their question.
Nov
2
comment Generalized additive models — who does research on them besides Simon Wood?
For example, see Bayesian Smoothing and Regression for Longitudinal, Spatial and Event History Data by Fahrmier & Kneib for a wide ranging coverage of the structure additive model approach.
Nov
2
comment Generalized additive models — who does research on them besides Simon Wood?
Thomas Kneib and Fabian Scheipl are two names that I am familiar with from this field and who promote a somewhat different way of fitting GAMs and related models. I get the impression that there is friendly "competition" between Simon Wood and these guys as I see Wood developing new ideas in papers & features in mgcv that are in "response" to the work of Kneib & Schiepl, and others. Knieb for example is one of the developers of BayesX which fits structured additive models & is somewhat different from Wood's penalized regression approach.
Oct
31
comment Why does the null deviance in glm.nb differ between models of the same response variable?
@Airone These things get tricky and I am not sure of the correct answer. Sounds like that would be a good question for Cross Validated, so I suggest you ask another question, linking to this one for some of the detail.
Oct
30
revised Choosing between LM and GLM for a log-transformed response variable
typos
Oct
30
answered Why does the null deviance in glm.nb differ between models of the same response variable?