49G 50G STAT Version 5 Multiple linear regression & Restricted model & Anova 1way

[Gil]

I just integrated now, automatically, in the matrix of the results :
- the VIFs (Variance Inflation Factors) &
- the CONDITION[/i] Indexes

As always, just copy the full Directory
in your phone appli EMU48 and enjoy your statistical analysis.

Gil

Version 3.8 fixes the Var coeff:
- Mean of Y calculated beforehand;
- Var coeff set as infinity when mean of Y = 0.
& explains better the use of 5 CF/SF in Version 3.8e.

Versions 4d... 5
include ANOVA-1 Way with automatic result with linear regression.

[SammysHP]

Thank you for your work! It might be easier to collect all posts of this tool in a single thread. Maybe a mod can merge the threads.

[Gil]

I agree, you are right.

I was not sure what was better: new thread for new version or old thread + reply with inclusion of new version of the former (old) program; apparently, the second option seems preferable.

Regards.

Gil

[rprosperi]

(01-17-2020 10:05 PM)Gil Wrote:  I agree, you are right.

I was not sure what was better: new thread for new version or old thread + reply with inclusion of new version of the former (old) program; apparently, the second option seems preferable.

Regards.

Gil

What seems to work best here is to extend the old thread, but also update your initial post (top of 1st page) with the latest version of the file, and any related notes about usage, changed specs., etc.

This minimizes the number of threads on the same topic, makes it easier for folks trying to find the thread on your topic, and by always updating the initial post, readers know where to find the latest files & notes without searching through the full thread.

[Gil]

Just for your information :

The programs themselves are to be found
in STAT/REG.

The documentation or notes are to be found
in STAT/NOTES.

Exact Input-Argument form for each of the 6 main programs fully explained, each time, at the beginning of the respective program (in form of a long string).

[Gil]


3.4 19.1.2020


Changed 
warning

message-
levels

for 
colinearity
:
VIF
10

CONDIT.IND
15&30


In last line of
RESUL of
C.REG & C.RESTR?:
[Screens 1.4&2.5]

unit ¦%¦
 added
to 
VARIAT
.
COEF


[rprosperi]

(01-18-2020 02:48 AM)rprosperi Wrote:  What seems to work best here is to extend the old thread, but also update your initial post (top of 1st page) with the latest version of the file, and any related notes about usage, changed specs., etc.

However, feel free to ignore that advice...

[Gil]

Just changed way of calculating VIF (Variance Inflation Factors) when multiple linear regression model is without intercept (same results now as with SAS or R statistical programs).

Besides that, the structure remains the same : probabilities, t, beta, covariances, (¹/²) (partial) correlations, test on beta's and on alternative model suppressing independent X variables, (adjusted) R2, multicolinearity (condition number), anova, Spearman’s coefficient and critical values, etc.

Explanations and examples still in the downloaded file and to be read only with the calculator as they were programs by themselves.

Commentaries welcome.

Regards,
Gil

[Gil]

New version 3.7 with a possible error in the variable name R2mod and its location.

Version 3.8 fixes the Var coeff:
- Mean of Y calculated beforehand;
- Var coeff set as infinity when mean of Y = 0.

Versions 4d-5
nclude now ANOVA-1Waywith the linear regression.

Regards,
Gil Campart

[Gil]

Version with .hp ending
(instead of the previous, confusing. DOC ending).

[Gil]

New version 5
including Analysis of variance ANOVA -1 Way
by the linear regression, with model automatically prepared for the user and appropriate conclusion about the equality of the groups means.

Suppose that, as shown below, you have 4 treatments and n=17 observations, with n1 (for 1st treatment) = 3, n2=4, n3=4 and n4=6.

To check the equalities of the 4 means
(mean1=1.7, mean2=2.8, mean3=4.125 and mean4=4.95), we can procede here very simply, just by one touch of key, running the special C.MODEL program (in the directory DATA ANOVA.1WAY), that automatically prepares and calculates/concludes everything.

For treatment 1, the results were
1.6
2.3
1.2

Go to Directory DATA ANOV.1WAY
Save these 3 stack values into the X1 (or other name) Matrix (dimension [3×1]) as follows:
{3 1} —>ARRY 'X1' STO

For treatment 2, the results were
2.5
3
2.8
2.9

Remain in Directory DATA ANOV.1WAY
Save these 4 stack values into the X2 (or other name) Matrix (dimension [4×1]) as follows:
{4 1} —>ARRY 'X2' STO

For treatment 3, the results were
3.5
4.2
4.3
4.5

Remain in Directory DATA ANOV.1WAY
Save these 4 stack values into the X3 (or other name) Matrix (dimension [4×1]) as follows:
{4 1} —>ARRY 'X3' STO

For treatment 4, the results were
4.1
3.9
4.6
5.2
5.9
6

Remain in Directory DATA ANOV.1WAY
Save these 6 stack values into the X4 (or other name) Matrix (dimension [6×1]) as follows:
{6 1} —>ARRY 'X4' STO

Remain in Directory DATA ANOV.1WAY
Now put the following list
(containing the appropriate 4 names that you created)
{X1 X2 X3 X4} in stack level 1

and, always in that directory, just press C.MODEL

Four results a, b, c & d will be automatically generated:
a) for the n (here 4) groups of treatments, creation of new n-1 (here 3) independent variables always called TR.1 TR.2... TR.n-1 (no TR.n!), here then TR.1 TR.2 & TR.3 (no TR.4!), and creation also of the dependent variable always called Y.TR that contains all the results of the n (here 4) treatments groups;
b) a "hidden" list for the linear model that will be used automatically, here
{1 TR.1 TR.2 TR.3 Y.TR} (with always the constant 1 as first element and then the variables TR.1 TR.2 TR.3 Y.TR specially adapted for the ANOVA-1Way Analysis with the linear regression);
c) the final visible output, ie the test result relative to the equality of the n groups means mu (here 4),
d) together with the detailed ANOVA table (with the corresponding probability).

It says that Fc(Calculated with the effective data of the experiments) in last column of ANOVA, line 2, = 20.54*
> Ftable(0.05, v1=3, v2=13), last column, line 3, = 3.41.
—> You should be very confident in rejecting the Ho Hypothesis that the 4 means** are equal at 5% level.

* The complex part, when editing that cell, is the corresponding calculated probability: 3.26*10^(-5)
or 3.26*10^(-3)% (here, a very, very tiny probability that we might however accept that at least two means mu — among the n groups/means, here then among the 4 groups/means — are equal, not knowing though which means mu_i might be considered as equal).

** The n means (relative to the last comparison test — last run of C. MODEL in directory DATA ANOV.1WAY — for the n groups means, here 4 groups/means) may always be found by pressing muANOVA (F-key on the right of ANOVA-Menu-Variable).

[Gil]

Version 5

With automatic ANOVA-1Way.