Recent content by Number Nine

When I say it's probably not what you want, I mean it's probably a bad idea, and it doesn't really have much statistical justification. All of your problems stem from the fact that the PO model isn't a good model. You don't use it to estimate CL and V because it doesn't provide trustworthy...

OP: You can do this pretty easily in R. Just write a function f(CL, V, F, ka) = \alpha E_\text{iv}(CL, V) + (1-\alpha) E_\text{po}(CL, V, F, ka) where E_\text{iv} is a function returning the sum of squared errors for the IV model given parameters CL, V, and E_\text{po} is the same thing for the...

The MCMC approach is almost certainly (and the FME documentation confirms this) fitting a Bayesian model. The approach your colleague is suggesting is, I think, similar to what I alluded to in a previous post, where you would derive a posterior distribution for the parameters in the first model...

I was confused about this as well. You can account for uncertainty in the parameters estimated in model one in a few ways (e.g. using their posterior distributions as priors in model two, or marginalizing over their uncertainty in model two), but none of them involve fixing them to exact values...

I'm very bored, and I think I have a good way to do this, so give me an hour or two and I'll get back to you with some R code that I think will do what you want. EDIT: Can you explain your "condition A" in more detail? If the PO data provides information about the parameters CL and V, it would...

A minor point: the "population standard deviation" (i.e. the square root of the sum of squared deviations from the mean, divided by n-1) is actually a biased estimate of the standard deviation. This follows from Jensen's inequality, since the square root is a concave function. It's fairly...

The "sample standard deviation" (the sum of squared deviations from the mean, divided by n) is a biased estimator of the population SD. It's actually very difficult to construct an unbiased estimator of the standard deviation, even for a normal distribution.

This is actually a bit of a tricky question. The simplest approach would be to just compare all possible models and select the best one, which would be 2^7 = 128 models in your case, which might take a while if you have to code them manually. If you want to try something a little more...

If you want to do prediction, why use AIC? Why not evaluate prediction error directly? Cross-validation is what you want. Just compare all the models you're interested in and choose the one with the best predictive performance. The AIC is actually asymptotically equivalent to leave-one-out cross...

Keep in mind that the KL-divergence is non-commutative, and different "orders" correspond to different objective functions (and different research questions). The way you're fitting it (that is, KL(Q||P), where Q is being fit to P) is trying to match regions of high density, and it does seem to...

What are you trying to do, exactly? Generally you wouldn't use linear regression for longitudinal data without some modifications.

As an aside, it should be said that there is really no reason to ever use a pie chart, and most people involved in any kind of data visualization despise them. That's not just a personal preference; they're objectively terrible at communicating information. It's almost impossible to directly...

It should be noted that neither stepwise regression nor fit indices like the AIC really protect against overfitting. Stepwise regression in particular is kind of a joke among statisticians, though it's still (unfortunately) popular in the social/biological sciences. If you're concerned about...

The central limit theorem will not guarantee this; nor will anything, really. Given an appropriately chosen regression model, the errors may or may not be normally distributed. If they're not, then you'll have to either transform the data in some way, or fit a regression model with a different...

"Little Rudin" is a classic, and is a good textbook for people who have a strong background in pure math, and are already familiar with the basics of real analysis. That said, I would absolutely not recommend it as an introduction. Rudin's proofs are very "nice", in the sense that they tend to...