You are on page 1of 8

Linear regression:

Analyze- regression- linear


Statistcs t------ tickdescriptives, model fit, r sqr

Written on sheet.
Significance test of correlation test by observing p value
P < 0.05
Correl is significant
Rsqr =
Eqn- refer coeff table

Std error from model summary table


If SE is <<<<<< ybar (DV mean)
Model is fit else not

Testing of slope
H0 = B1=0
H1= B1not 0
T critical value = tinv(0.05,n-k ) refer model summary for dof, nik = df2
Compare t of IV
Tstat< tcrit reject null

P value approach
Tdist( x ,df, 2) xt cal
Pvalue alpha reject/accept (TABLE)

Multiple Regression

Res obj
Analyze- regression- linear
Statistcs t------ tickdescriptives, model fit, r sqr
Write from below

Y = b + b1 x1 + b2 x2.. EQN
Std error from model summary
Compare with grand mean
R sqr check and inference
Adjusted R sqr check

Test of slope validity

H0 b1=b2=b3. = 0
H1 atleast 1 b is 0

Fstat = MSB/MSE from Table of spss


Large F model valid, small F not explained
Fcrit= FINV (0.05, df1, df2) from model summary table
Fstat > fcriti reject null
P value < 0.05 reject null
Check for multi collinearity : 1- 1/ r^2 if > 10 there is multicollinearity

ONE WAY ANOVA

Steps : Analyze- compare means- one way


Dependant list DV
Factor IV
Options- descriptive

Research obj
R sqr
Hypo u1= u2=un
There exist no diff in .
H1 : atleast 2 means differ
There exists diff.

F stat = MSB/MSE frm ANOVA table


Fcritical = Finv (0.05,df1, df2) from ANOVA table

F stat> f crit reject null

P value approach
Tdist( x ,df2, 2) xt cal
Pvalue alpha F criti reject/accept (TABLE)

Conclusion : enough eveidence to infer that no diff in.

2 WAY ANNOVA
AnalyzeGLM---Univariate
Put DV and IV
Fixed- IV
First is main, 2nd is blocking
Write :
Res Obj

We make 3 research objectives


RQ: - Is there a difference in mean no. of Jobs (Response/dependent) between
men and women?
Ans1: - There is a difference in mean no. of Jobs (Response/dependent) between
men and women (10.8 & 10.05)
Hypothesis 1(H0): - There is no difference in mean numbers of job between men
and women (Mu1=Mu2)
Hypothesis 2(H1): - There is a difference in mean numbers of job between men
and women (Mu1Mu2)
RQ2: - Is there a difference in mean no. of Jobs (Response/dependent) between
four levels of education?
Ans2: - There is a difference in mean no. of Jobs (Response/dependent) between
four levels of education

Hypothesis 3(H0): - There is no difference in mean numbers of job between levels


of education (Mu1=Mu2=Mu3=Mu4)
Hypothesis 4(H1): - There is a difference in mean numbers of job between levels
of education (Mu1Mu2Mu3M4) At least 2 means are different
RQ3: - Is there a difference in mean jobs between gender and education levels?
(interaction effect)
Or There is no interaction effect between.
There is interaction effect
H0: - There is no difference in mean jobs between gender and education levels
H1: - There is a difference in mean jobs between gender and education levels

Look in test bw subjects table


P value approach
P< 0.05 reject null
Conclusion:

Factor analysis :

Analyzedata reduction--- factor

All factors to variable shift


In descriptives- Tick KMO
N extraction --- tick Scree plot
In rotation tick varimax

OK

Interpretation :
In total Var exp table check cumm % in extraction column

80.xx % variation explained by y factors


In rotated comp matrix
Highlight vales > 0.7
Highlighted values are factor loading of variables
Name all components

Bartlets test
We want to prove the correlation between the components is insignificant
H0 : significant
H1 : insignificant

KMO table
Pvalue < 0.05
Reject null
Conclsion ;.

KMO test is for testing sampling adequacy


KMO is xx should be > 0.5 for FA to be done

Communalities : check in comm table


Comm for 1st variable : 0.72
72 % of variance of fisrt variable is explained by 3 factors

Discriminant :
Analyze--- classify----discriminant

Grp variable- to be foun in ques


Define range of grp var
IV- independent

Classify..tick summary
Statistic-----tick unstandardized
Tick means
Tick within grp correlation
Savetick discriminant score

Interpretation

Pooled Within-Groups Matrices

Age income yearsmarriage

Correlation Age 1.000 -.411 .388

income -.411 1.000 .011

yearsmarriage .388 .011 1.000

Since none of the prediction/IV has correlation coeff > 0.75, we can say that
there is no multicollinearity condition
, if its more than 0.75 or less than 0.75 correl coeff or between 0.75 and
-0.75(present), we can say absence of multi collinearity

Standardized Canonical
Discriminant Function
Coefficients

Function

Age .924
income .775
yearsmarriage .151

3 independent variables- age income and years are predicting risk, out of these 3
variables, Age and income are more robust out of the 3 (higher the value higher
prediction)
Please note absolute value of std coeff for each variable indicates its relative
importance( ignore sign)

Canonical Discriminant
Function Coefficients

Function

1
Age .246
income .000
yearsmarriage .085
(Constant) -10.003

Unstandardized coefficients

Linear discriminant eqn = y = -10+ 0.246age + 0.0008 Income + 0.085 years

Eigenvalues

Canonical
Function Eigenvalue % of Variance Cumulative % Correlation

1 2.136a 100.0 100.0 .825

a. First 1 canonical discriminant functions were used in the analysis.

Canonical Correl is found out between discriminant score and original grp classification.

The sq of canonical correl is 0.68 , ie 68% variance in discriminating model between low and
high risk customer is explained by age income and years.

Wilks' Lambda

Test of Function(s) Wilks' Lambda Chi-square df Sig.

1 .319 16.573 3 .001

Wilks lambda is about the discriminating power of the model, it lies between 0 1
Lower the value better is the model
The discriminant model is significant as p value < 0.05
1-0.001 * 100 = 99.9 ,

Functions at Group
Centroids

Function

Risklevel 1

low 1.378
high -1.378
Unstandardized
canonical discriminant
functions evaluated at
group means

Are co
1 we have actual prediction in the data set, we compared actual prediction, with
discriminant score prediction (actual vs predicted) in that we observed that, all matches
except number 7. There are total 18 observations , 17/18 obs = 94.4% are correctly predicted.

New data- age 40 , income- 25000, years- 15

You might also like