0

I have a data that I have used discriminant function analysis with. In the results, one variable has a standardized canonical function coefficient that is greater than 1.0. I didn't think these could be greater than 1 and I am not sure what to make of it.

user40530
  • 1
  • 2
  • 1
    how, exactly, are these being standardized? – Glen_b Feb 19 '14 at 02:10
  • I guess what I meant is it possible for standardized canonical correlation coefficients to be greater than 1 (i.e. is it indicative of something wrong)? – user40530 Feb 19 '14 at 02:40
  • I can't answer your question until I know in what what the canonical correlation coefficient was 'standardized'. Define 'standardized' in this context. – Glen_b Feb 19 '14 at 03:01
  • IV's were items from factor analysis that were created from self ratings on measures of different problem behaviors (e.g. stealing $50+, using drugs, ect). DV was having/not having ADHD. Does this help? Otherwise, where can I look for how they were standardized in the results? – user40530 Feb 19 '14 at 03:28
  • sorry, typo above, should read *in what way the correlation coefficient was standardized'. No, telling me what the variables were doesn't help. I need you to define the quantity your question relates to. How do I know - for certain - what you're asking about? – Glen_b Feb 19 '14 at 03:34
  • To clarify: the word 'standardized' might mean more than one thing. It's not clear to me which of a number of possible things it actually is in this context. What software, at least, are you using? – Glen_b Feb 19 '14 at 05:40
  • How much greater than 1.0 was the correlation coefficient? Small errors could be caused by floating point maths. That also depends on how many samples you have. – chippies Feb 19 '14 at 11:05
  • There is _no_ such a thing as standardized canonical correlation coefficient in discriminant analysis. Are you mixing "canonical correlation coef." with "standardized canonical function coef."? – ttnphns Feb 19 '14 at 17:20
  • I am running SPSS . The value I got was 1.174 for the canonical correlation coefficient (although there was another less critical analysis that went as high as 1.4 for that particular variable). SPSS labels this as standardized canonical correlation coefficient. The sample was 14,000+ – user40530 Feb 20 '14 at 01:53
  • I'm SPSS user and I state that there is no `standardized canonical correlation coefficient` Can you point me output table where you saw that name? – ttnphns Feb 20 '14 at 05:37
  • It's a table just above the structure matrix table. But let's assume it IS the canonical discriminant function correlation coefficient, what would be the answer to my question? – user40530 Feb 21 '14 at 01:52
  • Gosh. It is not "canonical discriminant function _correlation_ coefficient" it is _canonical discriminant function coefficient_. So please press "edit" button below you question and edit it accordingly. – ttnphns Feb 21 '14 at 10:17
  • In discriminant analysis, standardized coefficients can be above 1. They are not standardized in the sense to fit a 0-1 range. They are standardized just to make variables with different variances comparable. They are computed as shown in the last paragraph [here](http://stats.stackexchange.com/a/48859/3277). – ttnphns Feb 21 '14 at 11:01
  • Thank you for answering my question. I'm a relative novice in this kind of statistics so I'm sure you will remember that you were once learning too. – user40530 Feb 22 '14 at 01:23

1 Answers1

2

This link may be helpful: http://www.ssicentral.com/lisrel/techdocs/HowLargeCanaStandardizedCoefficientbe.pdf

Also see this reference: Deegan, J.Jr. (1978). On the Occurrence of Standardized Regression Coefficients Greater Than One. Educational and Psychological Measurement, 38, No. 4, 873-888.

"If there is a single predictor or multiple predictors that are uncorrelated, then the beta values will be confined to the bounds of (-1,1). However, if there are 2 or more predictors that are correlated, positively or negatively, then the beta values may exceed those bounds."

In other words, you probably have multicollinearity issues.

Silverfish
  • 20,678
  • 23
  • 92
  • 180
f x
  • 21
  • 3