Henry
Reputation
12,446
Top tag
Next privilege 15,000 Rep.
Protect questions
 Apr 26 comment 'Level' still seems periodic after Season Decomposition A couple of points to consider: (1) the range of "level" is about $0.05$ (and the daily fluctuation even smaller) compared with $0.25$ for "season1" and (2) you you might take averages of level for individual hours (ignoring days) to see if there is a much of a pattern. My guess is that the real daily seasonal pattern changed very slightly from beginning to end (especially in the final $8$ days) and this is causing some of the apparent pattern in "level" Apr 26 comment Filter only names of Cities from datasets Not a Cross Validated question. How would you know that Alert might be a geographical name? Apr 26 comment Estimator of true probability — understanding margin of error for very small probability It looks like it. Note that $\tilde{p}$ is slightly different to $\hat{p}$ Apr 26 comment Estimator of true probability — understanding margin of error for very small probability The issue is the line $\displaystyle s_p = \sqrt{ \frac {p \, (1-p) } {n} } \le \sqrt{ \frac {0.5 \times 0.5 } {n} } = \frac {1}{2 \, \sqrt{n}}$ related to the title of the article Checking whether a coin is fair. If instead $p$ is close to $0$ or $1$, then $s_p$ will be much smaller. You might want to read the Wikipedia article Binomial proportion confidence interval which has some other approaches Apr 25 comment What transformation should I use for a bimodal distribution? @M.Beausoleil - in that case you really do not want to remove the bimodality, as it will reduce the discrimination of your analysis Apr 24 comment Distribution of sum of squared normals, scalar form Apr 20 comment Making sense of Binominal GLM model Perhaps this is telling you that a polynomial fit is not a good idea Apr 20 comment Adding Normal Distributions Your "standard formula" for $Z$ assumes independence between scores on the two tests. Is that true in your example? Apr 20 comment Convergence in distribution of the following sequence of random variables If it converges, then it converges to a distribution with point probabilities at $0$ and $1$, and which has an expectation of $\frac{\alpha}{ \alpha + \beta}$, so that gives you something to aim for Apr 12 comment For a standard normal distribution, are p-values calculated using a $\Phi(-Z)$ or $\Phi(Z)$? $\left|\dfrac{X-\mu}{\sigma}\right|$ will be non-negative, with the consequence that $2\Phi\left(\left|\dfrac{X-\mu}{\sigma}\right| \right) \ge 1$ so is unlikely to give a probability Apr 11 comment unbiased estimate for household size Note that if a village had half its households size $2$ and half size $6$ (so three-quarters of the inhabitants of the village lived in households size $6$), and you asked everybody in the village (e.g. stopping them in the street) then the average response would be $5$, while if you knocked at each door and asked the person opening the door, the average response would be $4$ Apr 4 comment Why do a density plot and a rug plot seem to disagree? @RickHenderson: I added a chart with a narrower bandwidth and manual jitter Apr 4 comment In statistics, should I assume $\log$ to mean $\log_{10}$ or the natural logarithm $\ln$? ISO 31-11 seems to specify $\ln$ for $\log_e$, and leave an unadorned $\log$ undefined Apr 4 comment In statistics, should I assume $\log$ to mean $\log_{10}$ or the natural logarithm $\ln$? Since the graphs on pages 6 and 7 of Gale's paper show the original units on a log scale, and the calculations are aimed at the slope of a log-log relationship, i.e. $b$ in the expression $\log(N_r ) = a + b \log(r)$ which corresponds to $N_r=Ar^b$, it makes no practical difference in this case Apr 1 comment Is this the solution to the p-value problem? @Glen_b: Sorry - I missed the tag and my only excuse was that it was still March in my time zone when I read the message. whuber: The never-existent "Journal of the American Statistical Society" apparently gets half a million hits, and there is probably an interesting study in patterns of when it has been cited as such in research papers Mar 31 comment Is this the solution to the p-value problem? You do realise that this American Statistical Society is not a real organisation? (It was the name of the American Statistical Association from November 1839 to February 1840, but that hardly counts.) STOP is a joke Mar 22 comment What are the most misused statistics terms that we should care to correct? @ttnphns: some of us write generalised with an s Mar 15 comment Difference between gaussian and lognormal I cannot see any reason to expect a simple solution Mar 9 comment Does the theory fit the given data? I suspect the number of degrees of freedom is less than $3$ given the calculation of MLEs Mar 9 comment Does the theory fit the given data? Given the MLEs (which do not quite add up to $1$ due to rounding), the expected number type O should be $353 \times 0.580^2$