5,834 reputation
11035
bio website www54.sap.com/industries/…
location Switzerland
age 38
visits member for 3 years, 11 months
seen 2 days ago

During the day, I forecast sales at supermarkets, drugstores, furniture, perfume and other retailers and calculate order proposals. Lots of time series, with an emphasis on fast, automatic and robust data cleansing and forecasting - with some logistical optimization thrown in for good measure. I'm active in the International Institute of Forecasters and an Associate Editor for their practitioner-oriented journal Foresight.

At night, I switch hats and do inferential statistics for clinical and biological psychology.

I'm never bored. And I use R.


Aug
29
comment Identifying outliers for within-patient numerical data
What is the "ESD test"?
Aug
29
answered Confidence bands for difference of time series
Aug
29
comment Visually summarising a mess of directed line segments
You write that you have segments, but in the graph you actually seem to have a piecewise linear trajectory (i.e., segments are all joined end-to-end)? Is that so? If yes, you may want to edit this into your question, since it may influence possible answers (which I unfortunately have no idea about).
Aug
28
comment How are these negative values understood?
It appears that you are looking at Cohen's $d$ as a measure of effect size; at least, your definitions of "low", "medium" and "high" are the ones Cohen proposed for his $d$. As per Alexis' answer, you can simply flip signs for negative $d$, keeping the interpretation in mind. However, note that $d$ is not constrained to be below 1 - $d>1$ is perfectly possible!
Aug
28
revised Best way to represent data
removed "in a histogram"
Aug
28
answered Best way to represent data
Aug
28
comment Chi square test of association
Please don't just discretize continuous data.
Aug
28
comment Evaluating performance of machine learning models
This question might be better suited to Data Science.SE.com, since it's not really about statistics per se. Consider flagging it for migration if you don't get good answers here.
Aug
28
answered Good form to remove outliers?
Aug
27
comment Fitting a straight line to components of complex numbers
The phase wouldn't change, since you would scale both the real and the imaginary part, and the phase depends only on the ratio between the two. Essentially, you would project a point in the complex plane onto the unit circle along the half-line from the origin through that point. (I still don't think I really understand what you are trying to achieve, though... maybe you want to edit your question to add a few more details.)
Aug
27
comment Why the Breusch-Pagan rejects H0 on apparently non-heteroskedastic data?
You are right. The problem probably rather is that the -200 residuals cluster at low fitted values.
Aug
27
comment Fitting a straight line to components of complex numbers
Why do you want to subtract the fitted amplitude from the original data? This seems like subtracting apples from oranges. Essentially, you have two-dimensional original data, fit a model to the absolute values and then want to subtract the fitted one-dimensional values from the original two-dimensional values. (You could scale the original data by dividing by the fits.)
Aug
27
comment Why the Breusch-Pagan rejects H0 on apparently non-heteroskedastic data?
Well. Using a SD of 6 (to roughly match your 1st and 3rd quartile), R gives me a probability of seeing a single -200 of 6.35e-244. You seem to have at least 15 or 20 residuals of -200. It seems like your tails are simply way too heavy to be expected from a normal distribution. Combine that with your large N, and any normality test will likely reject. You may be better served with the question whether you really need normality at all, given your huge N.
Aug
27
comment Why the Breusch-Pagan rejects H0 on apparently non-heteroskedastic data?
You seem to have quite a high N, right? This may just be a consequence of the fact that with a sufficiently large N, any deviation from H0 will get significant. I'd assume that, for instance, the large number of residuals clustered around -200 should "not happen" under normality.
Aug
26
comment How to handle large .csv file in R?
Where exactly do you get the error messages? I would assume in the call to efp(). What is the output of str(data)?
Aug
25
comment How to eliminate dependent inputs?
Principal Components Analysis?
Aug
25
comment using decompose function for high frequency data
Are you looking for the stl() function in the stats package? If not, please give a few more details.
Aug
20
answered Holt-Winters and Abnormal termination in LNSRCH
Aug
20
answered How to blend multiple time series models?
Aug
10
comment Prediction using Support Vector (SV) method in R
Quite honestly, I wouldn't be using SVM for time series, anyway. I don't even see how to model trend or seasonality using SVM. This may be helpful.