Reputation
Next tag badge:
199/100 score
19/20 answers
Badges
12 207 415
Newest
 Nice Answer
Impact
~3.3m people reached

3h
reviewed No Action Needed R packages to calculate Effect Size
3h
reviewed Leave Closed Convergence in Probability to a Constant
3h
comment Can binary data be ordinal?
+1 "Barely plural" and "otiose to muse on" are priceless--and right to the point.
3h
comment New to R , would like some help regarding statistical Arbitrage using R
@gung That may be a little unfair: implementing any useful strategy and backtesting it could be a substantial amount of work. On the other hand, we're not the right place to ask about the existence of code to do that sort of thing. Jai, do you have a statistical question to ask?
3h
reviewed Close Trainning a Text Detection System
3h
reviewed Close Extract details info from regsubsets
3h
reviewed Close Find likelihood ratio test statistic
3h
reviewed Reviewed When the results do not support the research
3h
comment When the results do not support the research
I am confused by how you phrase this question. It begins "suppose that you have ... research" but ends "what did I do wrong." Evidently something changed in the telling. What are you trying to ask?
3h
reviewed Leave Closed Sales prediciton
3h
comment How do I distribute answers lacking geo information from a poll?
You should enter into this endeavor fully aware of the possibility that missingness could have a (strong) relationship to the region (and therefore the answers probably should not be uniformly or randomly assigned to the regions). Did you collect, or can you collect, any information bearing on this?
3h
comment New to R , would like some help regarding statistical Arbitrage using R
Please tell us how you have been "stuck." There is an abundant amount of information available about R in all kinds of media, so it's inconceivable that you would have been unable to determine its capabilities to do the statistical computing you describe.
5h
comment Optimising simulation parameters - is it possible to produce regression style diagnostics such as p values?
OK, but what is your null hypothesis? You are adjusting all parameters of the simulation to match the data as closely as possible; that's fine. But there's no testable hypothesis involved in that, nor is one even implied.
5h
comment Correlation between OLS estimators for intercept and slope
The formula is correct, but could you please explain what asymptotics you are using? After all, in many cases the correlation does not vanish--it stabilizes. Consider, e.g., an experiment in which $x_i$ are binary and suppose data are collected by alternating $x_i$ between $1$ and $0$. Then $\sum x_i = \sum x_i^2 \approx n/2$ and the correlation will always be close to $\sqrt{2}/2 \ne 0$, no matter how large $n$ becomes.
6h
comment Error of calculated X from measured Y (instrument calibration with linear regression)
EdM has already given a good answer. If you wish to research further, it may help to know this problem is often solved with "inverse regression." Google "inverse regression" calibration for links to good resources.
6h
comment Optimising simulation parameters - is it possible to produce regression style diagnostics such as p values?
Thank you--that's an interesting application. But what does "model fit" mean? What is the model to which you are comparing the simulated results?
7h
comment Definition of stationary distribution in continuous time markov chains
+1. Thank you--this question has long needed a useful, relevant answer.
7h
comment How to generate data for gaussian distributions in these 2 scenarios in R?
Although this question has a statistical component--it would be of interest to explain the distinction between the two scenarios--its emphasis on R has caused it to collect purely programming-oriented answers, making it off-topic on CV.
7h
comment Incresing number of observations
It's getting clearer, but the example does not match its description. It appears you don't want to consider all possible combinations of times: you only want to consider ordered pairs where time1 precedes time2. But in that case why have you not also included the pairs $(10,50)$, $(20,50)$, and $(30,50)$ in the data3 example?
7h
comment How would I go about using regression to estimate a change in direction or movement in n-space?
That is helpful; thank you. It still leaves us with a vague problem description. For instance, it's trivial to detect changes as you have defined them: when the mean vector in one year is not identical to the mean in the previous year, a change has occurred. Your use of statistical terminology, though, suggests you have in mind some kind of probability model and some form of hypothesis test--you may be looking for significant changes. Further clarification of these points would be welcome.