Date: Thu, 9 Nov 2000 13:35:56 +0000 Jeremy Miles "SPSSX(r) Discussion" Jeremy Miles Re: simple regression reasoning To: Jeff Mather text/plain; charset="us-ascii"; format=flowed

At 13:14 09/11/00, Jeff Mather wrote: >I have a dependent variable (total events) and two IV's (total nurses, >total beds). >I have computed 2 ratio's, nurse's per bed (mean 0.4) or beds per nurse >(mean 2.6). WHen I run a simple linear regression looking at about 50 >different hospitals, the R2 for beds per nurse regressed against events is >significant, but the reverse ratio (nurses per bed) is not. > >What is the easiest way to explain why one is significant and the other is >not?

It might be because you have transformed the data, thus changing the distribution.

Think of it as similar to rate and speed. Rate is hours per mile, speed is miles per hour.

You do two things when you convert: 1) you change the shape of the distribution, and 2) you change the linearity. Graph below shows the conversion x axis is rate, y-axis is speed.

| * | * | * | * | * ------------------

I would suggest that you have a good look at your residual plots and see if anything is happening w.r.t. outliers, distributions or linearity.

HTH,

JM

---------- Dr Jeremy Miles j.n.v.miles@derby.ac.uk Phone: 01332 592090, Fax: 01332 593131, Mobile: 07941 228018 Inst. of Behavioural Sciences, Derby University, Derby, DE22 3HL, UK http://ibs.derby.ac.uk/~jeremym / http://www.jeremymiles.co.uk

Back to: Top of message | Previous page | Main SPSSX-L page