Determining the Correct Percentiles to Recode Data

[Paul]

I have 9 variables of behaviour scored in hours per week of
involvement. All scores are positively skewed (higher frequencies for
0 and 1-3 hours per week). Can SPSS calculate the correct percentile
cutoffs to recode the data? - Say 0 to 4 scales or 0 to 5 scales of
categories for hours per week (e.g., 0=0, 1= .1 to 2.9 hrs., 2=3.0 to
6.9 hrs., 3=7.0+).
As it stands, I eyeballed the raw output of percent and cumulative
percent for each score of the 9 variables and then simply
decided/judged what percentile cut off (20% or 25%) would bring about a
more normal distribution of scores.
Thank you in advance.

[Art Kendall]

If i understand what you are asking, adding /ntiles=5 to the FREQUENCIES
procedure will give 5 groups of approximately equal size.

The distribution woulod then be flat (uniform) ratehr than normal.


Art

Post by Paul
I have 9 variables of behaviour scored in hours per week of
involvement. All scores are positively skewed (higher frequencies for
0 and 1-3 hours per week). Can SPSS calculate the correct percentile
cutoffs to recode the data? - Say 0 to 4 scales or 0 to 5 scales of
categories for hours per week (e.g., 0=0, 1= .1 to 2.9 hrs., 2=3.0 to
6.9 hrs., 3=7.0+).
As it stands, I eyeballed the raw output of percent and cumulative
percent for each score of the 9 variables and then simply
decided/judged what percentile cut off (20% or 25%) would bring about a
more normal distribution of scores.
Thank you in advance.

[JKPeck]

Besides the ntiles available from Frequencies or from RANK, you can use
the Visual Bander interactively to create an appropriate recode. (That
came with SPSS 13, if memory serves). It can compute cutpoints based
on equal width intervals, percentiles, or mean and standard deviations
as well as user-determined cutpoints based on a little histogram
display..

It is accessed from the Transform menu.

Regards,
Jon Peck
SPSS