- View SourceOriginal Question:
> Is there a spatial equivalent to the non-spatial t-test? I've got two

Replies:

> samples that I want to compare, but one is auto-correlated. I produced

> the auto-correlated sample using kriging, so I don't want to take out the

> spatial info.

From Carlos Carroll <carlos@...>> THe CRH modified t-test may be useful. See Clifford et al. 1989.

SOME EXCELLENT REFERENCES! CLIFFORD'S ARTICLE, "ASSESSING THE

> Biometrics 45:123-134, or for an application , see Thomson et al. 1996.

> Ecology 77:1698-1715.

SIGNIFICANCE OF THE CORRELATION BETWEEN TWO SPATIAL PROCESSES" IS A GREAT

RESOURCE.

ANOTHER GOOD REFERENCE (IN FACT, A RESPONSE TO CLIFFORD'S ARTICLE) IS:

DUTILLEUL, PIERRE. 1993. MODIFYING THE T TEST FOR ASSESSING THE

CORRELATION BETWEEN TWO SPATIAL PROCESSES. BIOMETRICS 49: 305-314.

From Philippe Aubry <paubry@...-lyon1.fr>> If only one sample is spatially auto-correlated there is no need for a

THIS IS NOTED IN CLIFFORD'S ARTICLE: WHEN ONE OF THE PROCESSES HAS NO

> correction of the classical t test so the p-value you will obtain will

> be statistically reliable (e.g. using a randomization test). But, if the

> two samples show spatial autocorrelation, the p-value calculated in a

> usual way (i.e. for data no spatially dependent) may be very eroneous,

> depending on the magnitude and sign of the autocorrelation in each

> sample. In such a case you need to use a corrected procedure in order to

> evaluate the p-value associated with the observed t statistic.

AUTOCORRELATION, THE STANDARD T-TEST PERFORMS AS WELL AS THE CRH MODIFIED

T-TEST.

From Marcia Gumpertz <gumpertz@...>> I'm not sure what you mean by saying that you produced the

I HAVEN'T PURSUED THIS POSSIBILITY AS I DON'T HAVE SAS.

> auto-correlated sample using kriging, but you can use SAS proc mixed to

> fit a regression or analysis of variance model with spatially correlated

> errors. Then you can do a Wald test (like a t-test) to compare two

> treatments.

From "William C. Thayer" <wcthaye@...>> Have you considered multivariate tests such as Hotelling's T^2? I am

DIDN'T PURSUE THIS POSSIBILITY EITHER AS IT DIDN'T SOUND LIKE A SPATIAL

> not sure that this test is appropriate (i.e. do the two sets of data

> represent two populations? Will the test give valid results when one

> set of data is correlated?). The test is designed to consider the

> correlation between the two sets of variables. Wether the two sets of

> data can be treated as two variables measured at the same location (I

> assume that is the case) is another issue that (I think) should be

> addressed.

TEST.

From Hossein Arsham <harsham@...>> You may like to look at

THIS REALLY DIDN'T HAVE WHAT I WAS LOOKING FOR.

> Statistical Analysis of Circular Data,

> by N. Fisher, Cambridge University Press, 1993.

> and references therein.

Thanks to the above for replying to my plea for help!

Sarah

--

*To post a message to the list, send it to ai-geostats@....

*As a general service to list users, please remember to post a summary

of any useful responses to your questions.

*To unsubscribe, send email to majordomo@... with no subject and

"unsubscribe ai-geostats" in the message body.

DO NOT SEND Subscribe/Unsubscribe requests to the list!