topic Re: ttest for groups in Statistical Procedures

ttest for groups

roohaa — Thu, 06 Jul 2017 15:53:20 GMT

I have a proc means results like this

Group1	Group2	Xmean	Xstd	N	Ymean	Ystd	Yn	X/Y	X-y
A	0	0.68	1.96	101	0.70	1.89	112	0.48	-0.02
B	0	0.62	1.99	202	0.58	1.87	175	0.54	0.04
C	0	0.56	2.15	101	0.52	1.92	154	0.54	0.04
D	0	0.56	2.53	147	0.62	2.66	1146	0.45	-0.06
A	0	0.62	2.03	202	0.62	1.97	122	0.50	0.00
B	0	0.59	1.97	223	0.61	2.03	4631	0.48	-0.02
C	0	0.55	2.12	202	0.50	2.18	202	0.55	0.05
D	0	0.73	2.29	202	0.64	2.25	101	0.57	0.09
A	0	0.62	2.20	255	0.66	2.25	1144	0.47	-0.04
B	0	0.66	2.44	202	0.59	2.43	101	0.56	0.07
C	0	0.71	2.64	202	0.67	2.63	635	0.53	0.04
D	0	0.68	2.75	202	0.57	2.76	101	0.60	0.11
A	1	0.62	1.66	80	0.57	1.67	77	0.54	0.05
B	1	0.58	1.64	80	0.56	1.64	79	0.51	0.02
C	1	0.54	1.69	80	0.50	1.67	77	0.54	0.04
D	1	0.55	2.28	80	0.48	2.28	77	0.58	0.08
A	1	0.57	1.67	80	0.56	1.68	78	0.51	0.01
B	1	0.58	1.61	80	0.62	1.99	101	0.47	-0.03
C	1	0.53	1.71	80	0.34	1.57	26	0.79	0.19
D	1	0.57	1.90	80	0.47	1.92	74	0.61	0.11
A	1	0.57	1.93	80	0.47	1.96	74	0.60	0.10
B	1	0.57	2.03	80	0.56	2.06	78	0.51	0.01
C	1	0.62	2.15	80	0.51	2.19	73	0.61	0.11
D	1	0.60	2.23	80	0.39	2.17	30	0.77	0.21

and need to run the ttset for group 2, something like this

Group2	Xm (xmean)	Ym (ymean)	xm/ym	xm-ym
0	0.63	0.61	0.52	0.03
1	0.58	1.88	0.59	0.07
group2(0)-group2(1)	0.06	-1.27	-0.06	-0.05
t-value	?	?	?	?

however I am not sure how to use proc ttest for this case. in this example, 0.63 is the x variable mean for group2=0 (x itself is mean from each group0*group1 classification (first table).

I calculated the dufferences for mean, ratio and the difference adnd need to calculate the t-stats. I am not sure if covariances are zero.

I want to go from first table to the second table in SAS (any version). I would appreciate your comments.

Thanks

Re: ttest for groups

Reeza — Thu, 06 Jul 2017 15:56:36 GMT

1. Do you by chance have the raw data?

2. Is this a meta analysis?

Re: ttest for groups

roohaa — Thu, 06 Jul 2017 15:59:53 GMT

Yes, I have the raw data, which are panel data and I created the first table using a proc means.
No it is not a meta study. I am comparing two groups asset returns.

Re: ttest for groups

roohaa — Mon, 10 Jul 2017 14:07:29 GMT

of course, there is a class option for the ttest procedure but I have a mean of raios with different n and I am not sure if mean ratio has normal/tstudent distribution. So I need to consider different sample sizze for which I have a ratio, for example 101 ons for ratio .48 and also may need taylor linearization for the ratio distribution.

Re: ttest for groups

Rick_SAS — Mon, 10 Jul 2017 14:54:38 GMT

You can use the TEST=RATIO option on the PROC TTEST statement to test for the ratio of means. See the doc for PROC TTEST. You'll probably want to graph the data to determine if DIST=NORMAL or DIST=LOGNORMAL is appropriate.

Re: ttest for groups

Reeza — Mon, 10 Jul 2017 15:05:44 GMT

If you have the raw data you may want to analyze the raw data rather than the summary statistics.

I suspect that may be a more complex model than a t-test, most likely a mixed model

Re: ttest for groups

StatsMan — Mon, 10 Jul 2017 17:57:26 GMT

I would follow Rick's suggestion above, using the raw data with PROC TTEST.

If you have other factors in the data or if your groups really are a factor in a model, then using PROC GLM or PROC MIXED may be appropriate.

Re: ttest for groups

roohaa — Mon, 10 Jul 2017 18:31:41 GMT

What about using log-normal distibution with TEST=RATIO in proc ttest? Do you think it is suitable?