problems with Durbin Watson in linear regressions with more than 500 cases

[Pelusa Orellana]

Hi, I've been having the following problem. I'm conducting linear
regressions in which I have more than 500 cases, and the Durbin Watson
test gives me different values each time I sort the data differently.
How can this be? How can I know which one is the real value?
thanks!
Pelusa Orellana
UNC Doctoral student
UNC Chapel Hill.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Peck, Jon]

The Durbin-Watson statistic is a measure of serial correlation of the residuals.  It only makes sense to look at it when your data are in a natural order.  If you don't have time series data, you probably don't care about it, but for time series data you need them in time order.

HTH,
Jon Peck

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Pelusa Orellana
Sent: Thursday, January 03, 2008 2:07 PM
To: [hidden email]
Subject: [SPSSX-L] problems with Durbin Watson in linear regressions with more than 500 cases

Hi, I've been having the following problem. I'm conducting linear
regressions in which I have more than 500 cases, and the Durbin Watson
test gives me different values each time I sort the data differently.
How can this be? How can I know which one is the real value?
thanks!
Pelusa Orellana
UNC Doctoral student
UNC Chapel Hill.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[David Greenberg]

If you think about the way the Durbin-Watson statistic is defined, it is not surprising that its value will change when you sort the data differently. Each value is the "real value" for that ordering of the data. One would wonder, though, why you are doing this sorting. The Durbin-Watson statistic is for time series data, in which the ordering of the cases is meaningful. Sorting the data in a manner that disrupts that ordering is not something you should be doing with time series data. If you don't have time series data (and there is no other intrinsic significance to the ordering of the cases) then there is no reason to compute the statistic. David Greenberg, Sociology Department, New York University

----- Original Message -----
From: Pelusa Orellana <[hidden email]>
Date: Thursday, January 3, 2008 5:32 pm
Subject: problems with Durbin Watson in linear regressions with more than 500 cases
To: [hidden email]

> Hi, I've been having the following problem. I'm conducting linear
> regressions in which I have more than 500 cases, and the Durbin Watson
> test gives me different values each time I sort the data differently.
> How can this be? How can I know which one is the real value?
> thanks!
> Pelusa Orellana
> UNC Doctoral student
> UNC Chapel Hill.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD