Interpretation of Proc TTEST output?

[genemroz]

Below is the output from the following  Proc TTEST:

Proc ttest data=want alpha=.001;
where station in('US0001','US0002');
Class station;
var slm;
run;

The upper limit for the 99.9% confidence limit of the mean for US0002 is larger than the lower limit for US0001.  This suggests to me that there is no statistically significant difference between the  two means at 99.9% confidence level.  Yet the lower confidence limit for the DIF(1-2) value is positive whereas I expected it to be negative suggesting a statistically significant difference at the 99.9% confidence limit.  How should I interpret these seemingly conflicting results?

Thanks,

Gene

The TTEST Procedure

Variable: SLM

Station	Method	N	Mean	Std Dev	Std Err	Minimum	Maximum
US0001		25	5.9781	0.0150	0.00299	5.9527	6.0116
US0002		19	5.9560	0.0129	0.00296	5.9413	5.9852
Diff (1-2)	Pooled		0.0221	0.0141	0.00430
Diff (1-2)	Satterthwaite		0.0221		0.00421

Station	Method	Mean	99.9% CL Mean		Std Dev	99.9% CL Std Dev
US0001		5.9781	5.9669	5.9893	0.0150	0.0100	0.0268
US0002		5.9560	5.9444	5.9676	0.0129	0.00822	0.0260
Diff (1-2)	Pooled	0.0221	0.00694	0.0373	0.0141	0.0103	0.0215
Diff (1-2)	Satterthwaite	0.0221	0.00723	0.0371

Method	Variances	DF	t Value	Pr > |t|
Pooled	Equal	42	5.15	<.0001
Satterthwaite	Unequal	41.245	5.26	<.0001

Equality of Variances
Method	Num DF	Den DF	F Value	Pr > F
Folded F	24	18	1.34	0.5299

[PaigeMiller · Accepted solution]

---

@genemroz wrote:

The upper limit for the 99.9% confidence limit of the mean for US0002 is larger than the lower limit for US0001.  This suggests to me that there is no statistically significant difference between the  two means at 99.9% confidence level. 

This is an incorrect interpretation/usage of confidence limits. It does not indicate statistical difference or lack of statistical difference. Overlapping confidence intervals is not equivalent to a T-test.

Yet the lower confidence limit for the DIF(1-2) value is positive whereas I expected it to be negative suggesting a statistically significant difference at the 99.9% confidence limit. 

Not sure why you expect DIF(1-2) to be negative. The T-test has p-value < 0.0001 indicating that the two means are statistically different at your chosen alpha level.

[genemroz]

"Overlapping confidence intervals is not equivalent to a T-test." Interesting....thanks for this insight.

[Ksharp]

It is dependent on your ALPHA=.
If you set alpha=0.1 or 0.05, you could get different result .