SPSSX Discussion

Time series equivalent for One-Sample TTest

Classic

List

Threaded

6 messages Options

E. Bernardo

Time series equivalent for One-Sample TTest

A time series monthly continuous variable ( X_t, t=1, 2, ..., 24) is to be tested if its mean is 0. Since it is time series data, I guess I cannot the one-sample TTest. Can anyone suggest an appropriate approach for this?

Thank you.

Eins

===================== 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

Maguin, Eugene

Re: Time series equivalent for One-Sample TTest

Yes, why not. If not already done so, you have to restructure the file to long format.

Gene Maguin

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of E. Bernardo
Sent: Wednesday, July 09, 2014 12:49 AM
To: [hidden email]
Subject: Time series equivalent for One-Sample TTest

Thank you.

Eins

Jon K Peck

Re: Time series equivalent for One-Sample TTest

Since it is time series data, the observations are likely not independent, so the usual t test would be invalid.

However, you can simply fit an ARIMA(0,0,0) model with TSMODEL, which will estimate the intercept and show a t value for it (if you ask for the coefficients in the output.

Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
phone: 720-342-5621

From: "Maguin, Eugene" <[hidden email]>
To: [hidden email],
Date: 07/09/2014 11:05 AM
Subject: Re: [SPSSX-L] Time series equivalent for One-Sample TTest
Sent by: "SPSSX(r) Discussion" <[hidden email]>

Yes, why not. If not already done so, you have to restructure the file to long format.
Gene Maguin

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of E. Bernardo
Sent: Wednesday, July 09, 2014 12:49 AM
To: [hidden email]
Subject: Time series equivalent for One-Sample TTest

A time series monthly continuous variable ( X_t, t=1, 2, ..., 24) is to be tested if its mean is 0. Since it is time series data, I guess I cannot the one-sample TTest. Can anyone suggest an appropriate approach for this?
Thank you.

Eins
===================== To manage your subscription to SPSSX-L, send a message to LISTSERV@... (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 LISTSERV@... (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

Rich Ulrich

Re: Time series equivalent for One-Sample TTest

In reply to this post by Maguin, Eugene

If the problem is not the file structure, then it would be the question
raised by auto-correlation and its consequence -- data points are not
independent. And that assumption matters.

In any case of positive auto-correlations, the one-sample t-test will
be "generous" -- Thus, if the result is *not* p < 0.05, you can conclude
that there is slight evidence to reject the null.

One sort of correction for correlated data uses a t-test with a reduction
of d.f. that depends on the size of the shared correlation (to get an
"effective N"). However, that is less applicable to a time series.

I think that for time series in general, you would need to be careful
about "cycles" : If you model a non-stationary sine wave, you want to
test a parameter after accounting for the cycle, and not the raw mean.

Two years, taken monthly, does give a suitable mean. If you can parameterize
the seasonal variation to leave independent residuals, the resulting test
on residuals might be useful. If I used that, I think I would consider both the
difference from zero and the annual trend.

How I would do that in SPSS would depend on what was available as
for modeling the seasonal trend, from these points or elsewhere.

--
Rich Ulrich

Date: Wed, 9 Jul 2014 13:12:35 +0000
From: [hidden email]
Subject: Re: Time series equivalent for One-Sample TTest
To: [hidden email]

Yes, why not. If not already done so, you have to restructure the file to long format.

Gene Maguin

Thank you.

Jon K Peck

Re: Time series equivalent for One-Sample TTest

In reply to this post by Jon K Peck

Oops. You need to allow for some dependence explicitly with TSMODEL to get a valid test. You might try ARIMA(1,0,1) for a start.

Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
phone: 720-342-5621

From: Jon K Peck/Chicago/IBM@IBMUS
To: [hidden email],
Date: 07/09/2014 11:54 AM
Subject: Re: [SPSSX-L] Time series equivalent for One-Sample TTest
Sent by: "SPSSX(r) Discussion" <[hidden email]>

E. Bernardo

Re: Time series equivalent for One-Sample TTest

Dear Jon Peck,

I have tried the ARIMA(0,0,0) TSModel and the result is the same with One-sample TTest.
Also, I have tried the ARIMA(1,0,1) TSMODEl and found out that the t values for the constant and MA are nonsignificant (p>.01), while for the AR is significant (p<.05).

Thank you for your comment, Jon. You answered my question. Thanks also for those who commented.

Eins

On Thursday, July 10, 2014 2:56 AM, Jon K Peck <[hidden email]> wrote:

Yes, why not. If not already done so, you have to restructure the file to long format.
Gene Maguin

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of E. Bernardo
Sent: Wednesday, July 09, 2014 12:49 AM
To: [hidden email]
Subject: Time series equivalent for One-Sample TTest

A time series monthly continuous variable ( X_t, t=1, 2, ..., 24) is to be tested if its mean is 0. Since it is time series data, I guess I cannot the one-sample TTest. Can anyone suggest an appropriate approach for this?
Thank you.

Eins
===================== 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