At 09:25 AM 2/8/01 -0600, jim clark wrote:
>Hi
>
>I'll take Don's points out of order.
the overriding problem here is that ... the way stats are done on packages,
etc. ... the output is highly abbreviated ... you get
MTB > ttest 5 c1
One-Sample T: C1
Test of mu = 5 vs mu not = 5
Variable N Mean StDev SE Mean
C1 20 -0.082 0.914 0.204
Variable 95.0% CI T P
C1 ( -0.510, 0.345) -24.88 0.000 <<< forget the 0 here
but, what does this EXACTLY mean?
MTB > regr c2 1 c1
Regression Analysis: C2 versus C1
The regression equation is
C2 = 0.316 + 0.368 C1
Predictor Coef SE Coef T P
Constant 0.3164 0.1957 1.62 0.123 <<<< ???? where does
it come from???
C1 0.3676 0.2189 1.68 0.110
S = 0.8715 R-Sq = 13.5% R-Sq(adj) = 8.7%
Analysis of Variance
Source DF SS MS F P
Regression 1 2.1428 2.1428 2.82 0.110 <<<< one or
two tailed????
Residual Error 18 13.6723 0.7596
Total 19 15.8151
Unusual Observations
Obs C1 C2 Fit SE Fit Residual St Resid
1 0.26 -1.617 0.412 0.209 -2.029 -2.40R
R denotes an observation with a large standardized residual
MTB > chis c50 c51
Chi-Square Test: C50, C51
Expected counts are printed below observed counts
C50 C51 Total
1 23 49 72
34.79 37.21
2 49 28 77
37.21 39.79
Total 72 77 149
Chi-Sq = 3.997 + 3.737 +
3.737 + 3.494 = 14.965
DF = 1, P-Value = 0.000 <<< again, forget the 0 but, one or two tailed?
MTB >
NONE of this output really says what is going on ... what the actual
statistical test is ... and the p value says nothing about 1/2 tails etc.
it is ASSUMED that the user KNOWS all this stuff ... which of course is a
mammoth MIS assumption to make
for the ttest output ... what minitab does (and i assume other packages)
... it takes the calculated t ... pretends it is BOTH + and - ... then
finds the area OUTside these boundaries on the relevant t distribution ..
and prints as p
for the chi square test ... it only goes ONE way ... so, finds the upper
tail end and reports it as p
for the ANOVA in the regression output ... same as for chi square
i guess it all boils down to users knowing what the test statistics are ...
how they work ... and how they related to using some particular
distribution like t or F or chisquare
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================