Cyan

1,415 reputation

bio	website
	location	Ottawa, Canada
	age
visits	member for	1 year, 3 months
	seen	Jun 16 at 19:52
stats	profile views	167

Corey Yanofsky, Ph.D.

I'm a biostatistician with an interest in financial statistics located in Ottawa, Canada. Philosophically, I'm a Bayesian of the Cox-Jaynes school. In practice I'm a statistical ecumenist whose role models are Michael I. Jordan and Andrew Gelman.

	1,415 reputation	bio	website		visits	member for	1 year, 3 months
39 badges		location	Ottawa, Canada		seen	Jun 16 at 19:52

summary answers questions tags badges favorites bounties reputation activity

191 Actions

suggestions reviews revisions comments badges posts accepts all

Jun 1	comment	What's the problem with model identifiability? Because if the model isn't identifiable, then the inferences are being driven by the prior. This isn't necessarily a bad thing (as your question points out), but it's always something that's important to know.
May 30	comment	What's the problem with model identifiability? Yes, it's valid -- as valid as Bayesian inference ever is, which is a matter of some dispute. (I say go for it.)
May 20	comment	Can Neyman-Pearson lemma apply to the case when simple null and alternative don't belong to the same family of distributions? @Glen_b and Tim: If "LR Test" is taken to mean a test whose statistic is a (possibly asymptotically) Chi-squared random variable, then the statement is true. That was a big deal in the era before powerful desktops that can simulate random samples from the null. Nowadays one can just code up the simulation and, by repeated sampling, get an arbitrarily accurate approximation of the p-value.
May 20	comment	Can Neyman-Pearson lemma apply to the case when simple null and alternative don't belong to the same family of distributions? "...you don't know the distribution of the likelihood ratio under the null, even asymptotically." That's not such a big concern in a world where you can code up a simulation under the null in less that 20 lines of R.
May 20	comment	Can Neyman-Pearson lemma apply to the case when simple null and alternative don't belong to the same family of distributions? As you say in the question, the proof makes no assumptions about the form of the two distributions. Trust the math.
May 10	revised	How to test if two samples are distributed from the same Gaussian process fixed a copypaste error
May 10	suggested	suggested edit on How to test if two samples are distributed from the same Gaussian process
May 6	comment	Are the posteriors “different”? How does one discuss the result? @rbatt, I don't have any references to hand that are directly on point; my thinking here basically echoes that of Andrew Gelman, author of Bayesian Data Analysis (3rd edition soon to be released).
May 6	answered	Are the posteriors “different”? How does one discuss the result?
May 2	comment	Bayesian analysis with histogram prior. Why draw simulations from the posterior? Bayesian specialist here. DJE's guess is 100% correct.
Apr 25	comment	Likelihood analysis for exponential distribution "In other words, where $\hat{\beta}_0$ is the MLE for $\beta_0$, will the profile likelihood be given by the following?" Depends what you mean by MLE. The hat notation is usually used for the global MLE. The profile likelihood needs a kind of "conditional" MLE given by $\tilde{\beta}_0 (\beta_1) = \beta_0 \, \mathrm{such \, that} \, \frac{\partial \log L}{\partial \beta_0} = 0,$ whereupon $L_p(\beta_1) = \log L(\tilde{\beta}_0(\beta_1),\beta_1).$
Apr 23	comment	Kalman- Bucy filter: prior mean change Changing the prior mean of $g$ is equivalent to changing the value of $\mu$.
Apr 23	answered	Do Bayes factors require multiple comparison correction?
Apr 17	revised	Understanding the derivation of an equation in LDA modeling improved formatting
Apr 17	suggested	suggested edit on Understanding the derivation of an equation in LDA modeling
Apr 12	revised	Bayesian parameter estimation of a Poisson process with change/no-change observations at irregular intervals Too tired last night, clearly.
Apr 12	revised	Bayesian parameter estimation of a Poisson process with change/no-change observations at irregular intervals fixed punctuation
Apr 12	revised	Bayesian parameter estimation of a Poisson process with change/no-change observations at irregular intervals had it right the first time
Apr 12	answered	Bayesian parameter estimation of a Poisson process with change/no-change observations at irregular intervals
Apr 10	comment	complicated model without a huge dataset Shrinkage just means moving parameter estimates toward zero; so from a certain point of view, model selection, i.e., setting some parameters of the full model to zero, is already a shrinkage approach. From this perspective, one ought to just use the full model, let whatever shrinkage approach one has chosen do its work, and then report when shrunken estimates cannot be distinguished from zero (using a bootstrap CI, perhaps).