At 02:19 PM 7/9/2004, Alexandre Cechin wrote:

>How can I apply a test to see if a response to an event is significantly >different from the historical event response? Assume that the dependent >variable is 0 or 1 (non-response and response). And to compare 2 different >trials, each one with its own event, to see if response to event number 1 >was different from event number 2? [Is it] better to use the raw data >(case by case) or to use aggregated/summarized data (with counts and % of >positive responses).

At 02:53 PM 7/9/2004, Hector Maletta wrote:

>Seems to me you should use the T-TEST procedure in SPSS. You use raw data >for that test.

Yes; T-TEST takes raw data only. (Summarized data loses standard-deviation information, which T-TEST needs.)

But you have a 0-1 dependent variable, and a 0-1 independent (one trial vs. history; one trial vs. another): CROSSTABS and CHI-SQUARE is appropriate.

Suppose your trials are numbered 1,2,... in variable TRIAL; and your dependent variable is RESPONSE, coded 0-1, as you say. This code is not tested; and you may want to change the CELLS display. It assumes raw (un-aggregated) data.

(You can run CROSSTABS with aggregated data that has counts of positive AND negative responses by trial, but it's more trouble. Code not given here.)

/* Compare trials 1 and 2 only */ TEMPORARY. SELECT IF ANY(TRIAL,1,2). CROSSTABS /TABLES= TRIAL BY RESPONSE /FORMAT= AVALUE TABLES /STATISTIC=CHISQ /CELLS= COUNT ROW .

/* Compare trial 1 with all others */ TEMPORARY. RECODE TRIAL (MISSING = SYSMIS) (1 = 1) (ELSE = 0). VALUE LABLES TRIAL 0 'Others'. CROSSTABS /TABLES= TRIAL BY RESPONSE /FORMAT= AVALUE TABLES /STATISTIC=CHISQ /CELLS= COUNT ROW .