Quartz | Level 8

## What kind of transformation procedure need to applied for this kind of distribution?

Hi All ~ I'm dealing with a contienous variable whose distrubution looks as shown in the two figures. Which tranformation procedure would you recommnend for this kind of distrubution?

25 REPLIES 25
Super User

## Re: What kind of transformation procedure need to applied for this kind of distribution?

Why are you transforming?

Quartz | Level 8

## Re: What kind of transformation procedure need to applied for this kind of distribution?

I'm using this as the dependent variable for my GLM analysis ~ which assumes the normal distrubution of outcome!

Super User

## Re: What kind of transformation procedure need to applied for this kind of distribution?

I'm using this as the dependent variable for my GLM analysis ~ which assumes the normal distrubution of outcome!

Logistic regression falls under GLM and a binary response is definitely not normal.

Super User

## Re: What kind of transformation procedure need to applied for this kind of distribution?

Your data is right bias. Try Log Normal distribution.

Super User

## Re: What kind of transformation procedure need to applied for this kind of distribution?

Box cox transformation using proc transreg may be an option

Quartz | Level 8

Quartz | Level 8

## Re: What kind of transformation procedure need to applied for this kind of distribution?

The distrubution shown here is the log tranformed values of orifinal values!

SAS Super FREQ

## Re: What kind of transformation procedure need to applied for this kind of distribution?

How did you create that panel of histogram + box plot + Q-Q plot?  Is that something you wrote with GTL or was it generated by a SAS procedure?

Super User

## Re: What kind of transformation procedure need to applied for this kind of distribution?

@Rick_SAS that looks like the output from proc univariate.

Quartz | Level 8

## Re: What kind of transformation procedure need to applied for this kind of distribution?

This is the output from PROC UNIVARIATE!

## Re: What kind of transformation procedure need to applied for this kind of distribution?

That is an interesting plot.  Even the log10 values show an extreme skew.  Is there a possibility that these are maximum values for something?  If so, you may want to look at a Weibull distribution or some other generalized extreme value distribution.  a ln(ln (x)) transformation might be useful.  I don't know if TRANSREG will hande this, though.

Steve Denham

Quartz | Level 8

## Re: What kind of transformation procedure need to applied for this kind of distribution?

I tried the ln(ln(x)) ~ The distribution almost looks the same with the reduced SD!
Quartz | Level 8

## Re: What kind of transformation procedure need to applied for this kind of distribution?

Here are the log(x) and log(log(x)) distrubution!

Super User

## Re: What kind of transformation procedure need to applied for this kind of distribution?

It's hard to see what your doing now, since the forum doesn't sort the responses. So no way to tell which reply goes with which post. In the future please just do one response. This isn't your fault - limitation of the forum.

Did you post your original distribution? Also, what is the variable, in laymans terms. Context can help with deciding what type of transformation to use, and there are certain standard transformation in diff industries.

Discussion stats
• 25 replies
• 1512 views
• 2 likes
• 5 in conversation