Distribution that works with positive and negative skewness

Hi,

I am trying to obtain the distribution type of a variable that changes over time. For each stage I have n number of data.

I use fitdist grouping by stage to get the distribution for each stage. The problem is that the distribution varies over time, being, in some stages, positive skewed (lognormal is the best distribution), in other stages symetrical (normal) and in other stages negative skewed (extreme value).

My question is if there is any distribution type that fits for all cases; and if there is not, how can I change the distribution type by stage as fitdist only allow one distribution type.

[Anel Ruiz]

Aditional, I would appreciate any alternative advise for analysing this kind of data. I thought of only use median and percentiles, but I'am still trying with distribution types.

Answer 1

[Jeff Miller]

The skew normal is in Cupid. The last parameter controls the amount and direction of skew. E.g.,

posSkew = SkewNor(10,1,2);
posSkew.PlotDens;

negSkew = SkewNor(10,1,-2);
negSkew.PlotDens;

Answer 2

[Image Analyst]

Did you see the skewed normal distribution:

https://en.wikipedia.org/wiki/Skew_normal_distribution

[Anel Ruiz]

Is this distribution in MatLab library? I can't find it.

[Image Analyst]

Not sure. Probably not. random() has a few dozen distributions -- would one of them work for you?