Hello Team:
I have 18 messages- rating scale (1-7). Data looks as follows: begin data ID M1 M2 M3 M4....M18 1 7 6 5 7 6 I need to do pair t-test of each message with rest of them i.e. M1 with M2 to M17 and then M2 to M1, M3 to M18 and so on. If I mannually write pairs it will be more than 600 combinations. Can you experts favour me by sharing a easier way out. Thanks, Manmit |
Hi Manmit,
Although I strongly disagree with the use of multiple paired t tests instead of a RM ANOVA (or, even, better, since your data are ordinal, Friedman tests followed by a convenient post-hoc method), if you *really* want multiple t tests, here you are: T-TEST PAIRS = m1 TO m18. This way it will run all the pairwise combinations. MS> I have 18 messages- rating scale (1-7). Data looks as follows: MS> begin data MS> ID M1 M2 M3 M4....M18 MS> 1 7 6 5 7 6 MS> I need to do pair t-test of each message with rest of them MS> i.e. M1 with M2 to M17 and then M2 to M1, M3 to M18 and so on. If MS> I mannually write pairs it will be more than 600 combinations Only 153, as a matter of fact. I insist you use Friedman (plus a post-hoc), or RM ANOVA (plus a convenient post-hoc). Regards, Dr. Marta García-Granero, PhD Statistician. |
In reply to this post by MANMIT SHRIMALI
Hi Manmit
I'd rather follow this exchange of messages here, if you don't mind... MS> Thanks Marta for the solution. I had written long syntax and MS> it did not work. Your syntax, short and sweet works perfectly MS> alright. One item I forgot to mention in my previous message is the problem of hight type I error. You're running 153 t-tests, the chance that at least one is significant by chance is really above 0.05 (i feel lazy right now, that's why I'm not computing it, but I know it will be over 99%: 1-(0.95^153)). An overall test (like a RM ANOVA or Friedman) can be useful to check if there at least one pair of variables that differ signifficantly. MS> Re: RM GLM, you raised very important point. As per my MS> understanding, we need to classify the dependent variable and MS> factors. In my given problem, I cannot classify any message as MS> dependent. Usually, when I have data in column I use pair-t test MS> and using indepdent t-test will not generate significance MS> different. E.g. if the data is in the following manner: MS> Begin data MS> Id Drug1 Drug2 Drug3 Drug4 MS> 1 7 7 7 7 (message followed...) Sorry, but I don't quite follow your reasoning. You can, and you should use a RM ANOVA before attempting all those paired t-tests (if it is not significant, then you needn't perform all those pairwise comparisons): * Simulated data (according to your layout) *. INPUT PROGRAM. - VECTOR m(18). - LOOP #I = 1 TO 20. - LOOP #J = 1 TO 18. - COMPUTE m(#J) = 1+TRUNC(UNIFORM(7)). - END LOOP. - END CASE. - END LOOP. - END FILE. END INPUT PROGRAM. FORMAT m1 TO m18 (F4). EXE. GLM m1 TO m18 /WSFACTOR = m 18 Simple /EMMEANS = TABLES(m) /WSDESIGN = m . Concerning the second point I mentioned: your data are not truly quantitative, but ordinal. You could try Friedman test NPAR TESTS /FRIEDMAN = m1 TO m18. -- Regards, Dr. Marta García-Granero,PhD mailto:[hidden email] Statistician --- "It is unwise to use a statistical procedure whose use one does not understand. SPSS syntax guide cannot supply this knowledge (nor this list), and it is certainly no substitute for the basic understanding of statistics and statistical thinking that is essential for the wise choice of methods and the correct interpretation of their results". (Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind) |
In reply to this post by MANMIT SHRIMALI
Thanks Marta. I am sorry I posted on personal message. I am using the list serve for very first time so was not aware of whether to reply your email or listserv. Thanks again for make me informed.
Manmit -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Marta García-Granero Sent: Wednesday, July 26, 2006 2:23 PM To: [hidden email] Subject: Re: pair t test for multiple variables Hi Manmit I'd rather follow this exchange of messages here, if you don't mind... MS> Thanks Marta for the solution. I had written long syntax and MS> it did not work. Your syntax, short and sweet works perfectly MS> alright. One item I forgot to mention in my previous message is the problem of hight type I error. You're running 153 t-tests, the chance that at least one is significant by chance is really above 0.05 (i feel lazy right now, that's why I'm not computing it, but I know it will be over 99%: 1-(0.95^153)). An overall test (like a RM ANOVA or Friedman) can be useful to check if there at least one pair of variables that differ signifficantly. MS> Re: RM GLM, you raised very important point. As per my MS> understanding, we need to classify the dependent variable and MS> factors. In my given problem, I cannot classify any message as MS> dependent. Usually, when I have data in column I use pair-t test MS> and using indepdent t-test will not generate significance MS> different. E.g. if the data is in the following manner: MS> Begin data MS> Id Drug1 Drug2 Drug3 Drug4 MS> 1 7 7 7 7 (message followed...) Sorry, but I don't quite follow your reasoning. You can, and you should use a RM ANOVA before attempting all those paired t-tests (if it is not significant, then you needn't perform all those pairwise comparisons): * Simulated data (according to your layout) *. INPUT PROGRAM. - VECTOR m(18). - LOOP #I = 1 TO 20. - LOOP #J = 1 TO 18. - COMPUTE m(#J) = 1+TRUNC(UNIFORM(7)). - END LOOP. - END CASE. - END LOOP. - END FILE. END INPUT PROGRAM. FORMAT m1 TO m18 (F4). EXE. GLM m1 TO m18 /WSFACTOR = m 18 Simple /EMMEANS = TABLES(m) /WSDESIGN = m . Concerning the second point I mentioned: your data are not truly quantitative, but ordinal. You could try Friedman test NPAR TESTS /FRIEDMAN = m1 TO m18. -- Regards, Dr. Marta García-Granero,PhD mailto:[hidden email] Statistician --- "It is unwise to use a statistical procedure whose use one does not understand. SPSS syntax guide cannot supply this knowledge (nor this list), and it is certainly no substitute for the basic understanding of statistics and statistical thinking that is essential for the wise choice of methods and the correct interpretation of their results". (Adapted from WinPepi manual - I'm sure Joe Abrahmson will not mind) |
Free forum by Nabble | Edit this page |