You are on page 1of 22

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


Assignment #2
You have received a data set from an employer. This employer is a high-stakes
researcher who wants several questions asked based on the data provided to you. The
researcher does not have any specific hypotheses about the data and leaves that decision
up to you.
Your first task is to prepare the data for analysis, and then,
1. Formulate a research question that utilizes a multiple regression and test the
model.
2. Formulate a research question using MANOVA based on the data and test your
hypothesis. Use ad-hoc contrasts and conduct a post-hoc analysis.
3. Do a discriminant analysis on the data and inform the researcher about the
results.
PART I: Data Analysis
The purpose of this statistical report is to examine the relationships, if any,
between hours of math homework, teacher support and math confidence. The following
is a thorough description and analysis of data collected from a group of 250 students. It is
expected that hours of math homework and teacher support will contribute to the level of
math confidence in students. It is my hypothesis that the variables of Teacher support and
Hours of homework will be correlated with increased confidence in math ability.
Research questions were developed from the hypothesis and include; which variable
contributes the most to confidence in math ability? Are there between-group differences
in the variables? What is the between group variability?
To examine the data collected, Data Screening was conducted to ensure that all
data is appropriate for usage in the subsequent statistical analyses. Following that, a
Multiple Regression, a MANOVA and a Discriminant Functional Analysis were
performed in order to examine the relationship between the variables noted above.
Data Screening
Prior to analysis, the data was screened to determine if the model met the usual
statistical assumptions such as linearity, error distribution and variance. It appears that
there are no out of range values as there were no extreme outliers in the data set. There
were no missing or invalid scores, and the total sample is considered valid (Table A1).
Higher variance may exist in the Student confidence variable as the standard deviation

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


and mean are higher than the other variables (Table A1). Possible univariate outliers
were indicated in hours of homework (Table A3) that should be further checked for
accuracy of data entry. However, as we are not transforming or deleting any data
collected, outliers will simply be noted.
The scatterplots indicate a linear relationship and equality of variance for the
variables. Skewness and kurtosis values for the variables Student confidence and Teacher
support are close to zero, indicating normal distribution; therefore we are able to assume
multivariate normality with these variables. However, skewness and kurtosis are
considered significant for Hours of math homework, as noted in Table A2. Multivariate
outliers are addressed within the multiple regression.
No significant correlation was noted between Confidence and Teacher support.
Student confidence was correlated most highly with Hours of homework at . Therefore,
these variables may not be independently accounting for variability in the model. These
variables will be considered separately to either confirm the assumption of
homoscedasticity or account for the lack thereof.
Table A1
Statistics

Valid
Missing

Mean
Std. Deviation

Graph A1
Univariate Outliers

Graph A2

confidence in ability

hours of math

to do math

homework per month

teacher math support

250

250

250

41.4483

12.5334

27.5500

10.83725

2.95483

5.25000

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics

Graph A3

Graph A4

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


Table A2
Statistics
hours of math

Valid

confidence in

homework per

teacher math

ability to do math

month

support

250

250

250

Skewness

.222

.940

-.027

Std. Error of Skewness

.154

.154

.154

-.376

2.994

-.512

.307

.307

.307

Missing

Kurtosis
Std. Error of Kurtosis

Graph A5

Graph A7

Graph A6

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


Table A3
hours of math

hours of math homework

Pearson Correlation

per month

Sig. (2-tailed)
N

teacher math support

Pearson Correlation

confidence in

homework per

teacher math

ability to do

month

support

math

-.033

.429**

.599

.000

250

250

250

-.033

.053

Sig. (2-tailed)

.599

250

250

250

**

.053

.429

.406

confidence in ability to do

Pearson Correlation

math

Sig. (2-tailed)

.000

.406

250

250

250

Multiple Regression
Do the number of hours spent completing math homework and a higher level of
teacher support positively influence the level of student confidence in their math
abilities? To run the regression the standard enter method was used for entering the
independent variables, to see how they contribute together for the dependent variable of
Confidence in ability to do math as the model (Model 1). An examination of the
descriptive statistics of the variables initially shows some difference in variability
between confidence in ability to do math and the independent variables of Hours of math
homework and Teacher math support. The standard deviation of 10.84 for Confidence in
math ability indicates that there may be unequal variance. However, a large sample size
of 250 participants should be sufficient to account for any violations to assumptions.
Correlations between the independent variables are low, ranging from -.033 to
.300 which shows low collinearity. Correlations between the dependent variable
(Confidence in ability to do math) and independent variables are illustrated in Table B2
and are low/moderate at .429 for hours math homework per month. A low correlation at
.053 for teacher math support and math confidence is demonstrated. The highest
correlation is between hours of math homework and confidence in ability to do math at
.429.

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


An examination of the model summary (Table B4) shows that the model accounts
for 20% of variability in the dependent variable, and that the residuals are most likely not
auto-correlated as the Durbin-Watson is within an acceptable level of 1.930. The effect
size is considered moderate at .188 which indicates that the independent variables do
somewhat predict variability in confidence in math ability.
The ANOVA showed that the independent variables together do significantly
predict variability in confidence in ability to do math, with an F score of 28.678 (Table
B3). The standardized coefficients account for approximately 42% for Hours of
homework and10% for Teacher support. In other words, Hours of homework is
significant predictor whereas Teacher support is not significant. The collinearity
statistics overall demonstrate that the model is sufficient as the variables are adequately
independent with tolerance ranging from .999, and variance inflation levels at 1.001.
Tables B6 and B7 are Mahalanobis Distance and Cooks distance scores. The
average Mahal. Distance is 2.988 which is within the critical value. However, the range
shows the minimum value as .026, the maximum as 19.062, and the critical value for the
model is 18.42 (Stevens, 2009). The maximum of 19.062 was outside the critical range
and needed to be checked by looking at the Cooks distance and graphically for
multivariate outliers. The Cooks Distance values are all below one (minimum .000,
maximum .221, and mean .005) and are within the acceptable range indicating that there
are probably not multivariate outliers as illustrated in the histogram (Table B7), the
normal P-P plot of regression standardized residual (Table B8), and the scatterplot (Table
B9). These graphs show that the residuals are normally distributed with no influential
outliers distorting the accuracy of the regression. Therefore, the model meets the
assumptions of normality of errors, normal distribution, linear relationship, and equality
of variance.
The mulitple regression demonstrated that the one of the variables does
significantly account for variability in confidence, (hours of math homework) which
provides some support for the original hypothesis as well as answering the question of
which variables may contribute to confidence and self-efficacy.

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


Table B1
Descriptive Statistics
Mean

Std. Deviation

confidence in ability to do math

41.4483

10.83725

250

hours of math homework per month

12.5334

2.95483

250

teacher math support

27.5500

5.25000

250

Table B2
Correlations

Pearson

confidence in ability to do

Correlation

math

confidence in ability to

hours of math homework

teacher math

do math

per month

support

1.000

.429

.053

.429

1.000

-.033

.053

-.033

1.000

.000

.203

.000

.300

teacher math support

.203

.300

confidence in ability to do

250

250

250

250

250

250

250

250

250

hours of math homework


per month
teacher math support
Sig. (1-tailed)

confidence in ability to do
math
hours of math homework
per month

math
hours of math homework
per month
teacher math support

Table B3
ANOVAb
Model
1

Sum of Squares
Regression

df

Mean Square

5511.061

2755.531

Residual

23732.963

247

96.085

Total

29244.024

a. Predictors: (Constant), teacher math support, hours of math homework per month
b. Dependent Variable: confidence in ability to do math

Sig.
28.678

.000a

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


Table B4
Model Summaryb
Change Statistics

Model

Adjusted R

Std. Error of the

R Square

Square

Square

Estimate

Change

Change

df1 df2

28.678

2 247

.434a

.188

.182

9.80229

.188

Sig. F

Durbin-

Change

Watson

.000

a. Predictors: (Constant), teacher math support, hours of math homework per month
b. Dependent Variable: confidence in ability to do math

Table B5
Coefficientsa
Unstandardized

Standardized

Coefficients

Coefficients

Collinearity
Correlations

Std.
Model

1 (Constant)
hours of math

Error

Statistics

ZeroBeta

Sig.

order

Partial Part Tolerance

VIF

17.812

4.307

4.136 .000

1.581

.210

.431 7.517 .000

.429

.431 .431

.999 1.001

.139

.118

.067 1.170 .243

.053

.074 .067

.999 1.001

homework per
month
teacher math
support
a. Dependent Variable: confidence in ability to do math

Table B6
Statistics
Mahalanobis Distance
N

Valid
Missing

Skewness
Std. Error of Skewness
Kurtosis
Std. Error of Kurtosis

250
0
3.445
.154
14.743
.307

1.930

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


Table B7
Descriptive Statistics
N

Minimum

Cook's Distance

250

Valid N (listwise)

250

Maximum

.00000

.22139

Mean

Std. Deviation

.0046846

.01599370

Table B8
Collinearity Diagnosticsa
Variance Proportions
Condition
Model Dimension Eigenvalue
1

Index

(Constant)

hours of math homework per

teacher math

month

support

2.940

1.000

.00

.01

.00

.046

7.996

.01

.67

.30

.014

14.745

.99

.33

.69

a. Dependent Variable: confidence in ability to do math

Graph B1

Graph B2

Leora Fisher

10

EDPS 612.02 - Psychological Measurement and Statistics


Graph B3

MANOVA
Research questions that are addressed by the MANOVA involve determining the
differences in the variables contributing to increased confidence in ability to do math, and
any group differences between boys and girls between these variables. The grouping data
shows that we have groups of similar size that contribute to the overall robustness of the
results (Table C1). Boxs Test of Equality of Covariance Matrices was .121, meaning
that we can assume that covariances are equal as non-significance was found (Table C2).
Bartletts Test of Sphericity shows that the variables are correlated enough to use in the
multivariate analysis (Table C3). In addition, Wilks Lambda was significant at .949,
which indicates the necessity of further analysis. Equal variance across groups can be
assumed as non-significance was found in the Levenes Test. (Table C5). The other
variables were found to be non-significant for group differences. There is low correlation
between most variables as seen in the Residual SSCP Matrix (Table C7). The profile
plots show that estimated marginal means of all the variables are within a small range
which means that there are no suspected significant group differences.
A series of post hoc tests were run with the result that no significant difference
was found with any of the variables between groups (Table C9). The Tukey HSD
homogeneous subsets also show that there are no significant differences between groups
on the variables. Therefore, we can assume that there are no statistically significant
differences in the variables contributing to confidence in math ability, and no large group
differences exist in the sample. The correlation between hours of homework and
confidence in ability, and the influence of hours of homework on the groupings are the

Leora Fisher

11

EDPS 612.02 - Psychological Measurement and Statistics


only areas that were indicated as having low/moderate significance in the MANOVA
conducted. Although Graphs C1-3 appear significant, a close visual examination shows
that the group differences are low-moderate at most.
Table C1
Between-Subjects Factors
N
grouping variable

1.00

88

2.00

90

3.00

72

Table C2
Box's Test of Equality of Covariance
Matricesa
Box's M

18.142

1.485

df1

12

df2

269475.410

Sig.

.121

Tests the null hypothesis that the


observed covariance matrices of the
dependent variables are equal across
groups.
a. Design: Intercept + group

Table C3
Bartlett's Test of Sphericitya
Likelihood Ratio
Approx. Chi-Square

.000
432.980

df
Sig.
Tests the null hypothesis that the residual
covariance matrix is proportional to an identity
matrix.
a. Design: Intercept + group

5
.000

Leora Fisher

12

EDPS 612.02 - Psychological Measurement and Statistics


Table C4
Multivariate Test Results
Value
Pillai's trace

.051

Hypothesis df Error df Sig. Partial Eta Squared

2.155

6.000 492.000 .046

.026

6.000 490.000 .046

.026

2.158

6.000 488.000 .046

.026

3.000 246.000 .016

.041

Wilks' lambda

.949 2.157

Hotelling's trace

.053

Roy's largest root

.043 3.514

a. Exact statistic
b. The statistic is an upper bound on F that yields a lower bound on the significance level.

Table C5
Levene's Test of Equality of Error Variancesa
F
confidence in ability to do math
hours of math homework per month
teacher math support

df1

df2

Sig.

.728

247

.484

1.331

247

.266

.849

247

.429

Tests the null hypothesis that the error variance of the dependent variable is equal across groups.
a. Design: Intercept + group

Table C6
Tests of Between-Subjects Effects
Type III Sum of
Source

Dependent Variable

Corrected

confidence in ability to

Model

do math
hours of math homework

Mean

Squares

df

Square

Partial Eta
F

Sig.

Squared

292.824

2.524 .082

.020

49.754b

24.877

2.893 .057

.023

130.197c

65.098

2.388 .094

.019

424419.207

424419.207 3657.972 .000

.937

38731.183

38731.183 4503.472 .000

.948

187428.467

187428.467 6875.948 .000

.965

585.648

585.648

per month
teacher math support
Intercept

confidence in ability to
do math
hours of math homework
per month
teacher math support

group

confidence in ability to
do math

292.824

2.524 .082

.020

Leora Fisher

13

EDPS 612.02 - Psychological Measurement and Statistics


hours of math homework

49.754

24.877

2.893 .057

.023

130.197

65.098

2.388 .094

.019

28658.376 247

116.026

2124.273 247

8.600

6732.866 247

27.259

per month
teacher math support
Error

confidence in ability to
do math
hours of math homework
per month
teacher math support

Total

confidence in ability to

458735.401 250

do math
hours of math homework

41445.548 250

per month
teacher math support
Corrected

confidence in ability to

Total

do math

196613.688 250
29244.024 249

hours of math homework

2174.026 249

per month
teacher math support

6863.063 249

a. R Squared = .020 (Adjusted R Squared = .012)


b. R Squared = .023 (Adjusted R Squared = .015)
c. R Squared = .019 (Adjusted R Squared = .011)

Table C7
Residual SSCP Matrix

Sum-of-Squares and

confidence in ability to

Cross-Products

do math

confidence in ability

hours of math

teacher math

to do math

homework per month

support

28658.376

3257.289

959.623

3257.289

2124.273

-85.712

teacher math support

959.623

-85.712

6732.866

confidence in ability to

116.026

13.187

3.885

13.187

8.600

-.347

3.885

-.347

27.259

hours of math
homework per month

Covariance

do math
hours of math
homework per month
teacher math support

Leora Fisher

14

EDPS 612.02 - Psychological Measurement and Statistics


Correlation

confidence in ability to

1.000

.417

.069

.417

1.000

-.023

.069

-.023

1.000

do math
hours of math
homework per month
teacher math support
Based on Type III Sum of Squares

Table C8
Contrast Results (K Matrix)
Dependent Variable

grouping variable Helmert Contrast


Level 1 vs.

Contrast Estimate

Later

Hypothesized Value

confidence in ability

hours of math

teacher math

to do math

homework per month

support

3.137

.932

-.930

Difference (Estimate - Hypothesized)

3.137

.932

-.930

Std. Error

1.430

.389

.693

.029

.017

.181

.321

.165

-2.295

5.952

1.699

.435

-.574

.178

1.348

Difference (Estimate - Hypothesized)

-.574

.178

1.348

Std. Error

1.703

.464

.826

.737

.701

.104

-3.928

-.735

-.278

2.781

1.092

2.974

Sig.
95% Confidence Interval

Lower

for Difference

Bound
Upper
Bound

Level 2 vs.

Contrast Estimate

Level 3

Hypothesized Value

Sig.
95% Confidence Interval

Lower

for Difference

Bound
Upper
Bound

Leora Fisher

15

EDPS 612.02 - Psychological Measurement and Statistics


Table C9
Multiple Comparisons
Tukey HSD
95% Confidence
Interval

Dependent Variable

(I) grouping

(J) grouping

variable

variable

Mean

Std.

Difference (I-J)

Error

Sig.

Lower

Upper

Bound

Bound

confidence in ability to 1.00

2.00

3.4234

1.61482 .088

-.3842

7.2311

do math

3.00

2.8498

1.71171 .221

-1.1863

6.8859

1.00

-3.4234

1.61482 .088

-7.2311

.3842

3.00

-.5736

1.70313 .939

-4.5895

3.4423

1.00

-2.8498

1.71171 .221

-6.8859

1.1863

2.00

.5736

1.70313 .939

-3.4423

4.5895

2.00

.8428

.43965 .136

-.1939

1.8794

3.00

1.0211

.46602 .075

-.0777

2.1200

1.00

-.8428

.43965 .136

-1.8794

.1939

3.00

.1784

.46369 .922

-.9150

1.2717

1.00

-1.0211

.46602 .075

-2.1200

.0777

2.00

-.1784

.46369 .922

-1.2717

.9150

2.00

-1.6036

.78271 .103

-3.4492

.2420

3.00

-.2560

.82967 .949

-2.2123

1.7003

1.00

1.6036

.78271 .103

-.2420

3.4492

3.00

1.3476

.82551 .234

-.5989

3.2941

1.00

.2560

.82967 .949

-1.7003

2.2123

2.00

-1.3476

.82551 .234

-3.2941

.5989

2.00

3.00

hours of math

1.00

homework per month


2.00

3.00

teacher math support

1.00

2.00

3.00

Based on observed means.


The error term is Mean Square(Error) = 27.259.

Leora Fisher

EDPS 612.02 - Psychological Measurement and Statistics


Graph C1

Graph C2

Graph C3

16

Leora Fisher

17

EDPS 612.02 - Psychological Measurement and Statistics


Discriminant Function Analysis
A descriptive discriminant analysis was conducted to follow up the MANOVA to
further examine which variables contributed most to group differences. As the variables
Hours of math homework and Math confidence seem to be the most highly correlated,
those will be will be addressed in the DFA.
The DFA group statistics table indicates that difference between groups overall
are small upon examination of means and standard deviations (Table D1) and we can
assume that there is relative equality of variance. The standard deviations are relatively
low; therefore we can infer that the standard error is low. In Table D2, the F scores do
not show significance for any of the variables. Again Hours of homework is approaching
significance at .057, but it does not meet strict statistical criteria. The results of the
analysis of the covariance matrices indicate that the covariance matrices are equal (Table
D3). The log determinants are close to one another, and the test results are not significant
at .121 (Table D4).
Canonical discriminant functions were calculated. Eigenvalues are low at .043
and .010, and the canonical correlation accounts for only a small percentage of variance.
Therefore only a very small amount of variability can be accounted by discriminant
function. However, Wilks Lambda does indicate significance at the .046 level indicating
that one function (function 1) accounts for the variability (Table D6). The standardized
canonical discriminant function coefficients show that Hours of homework and Math
confidence are more important predictors within-group differences (Table D7). The
differences are illustrated in the structure matrix (Table D8) in which Hours of homework
(.655) and Math confidence (.681) are correlated to discriminant function 1. Math
confidence and Hours of homework do predict some variability between groups, whereas
Teacher support does not. Therefore, the assumption is that the variables of hours of
homework and confidence in ability are more influential. The canonical discriminant
function coefficients show the influence of Hours of homework and Math confidence,
which could be used for developing areas of future study. In relation to our research
questions and hypothesis, the variables that contribute to multivariate group differences
are Hours of homework and Math confidence.

Leora Fisher

18

EDPS 612.02 - Psychological Measurement and Statistics


Table D1
Group Statistics
Valid N (listwise)
grouping variable
1.00

2.00

3.00

Total

Mean

Std. Deviation

Unweighted

Weighted

confidence in ability to do math

43.5015

11.00970

88

88.000

hours of math homework per month

13.1309

3.48480

88

88.000

teacher math support

26.8990

5.22980

88

88.000

confidence in ability to do math

40.0781

10.05532

90

90.000

hours of math homework per month

12.2881

2.47349

90

90.000

teacher math support

28.5026

4.86458

90

90.000

confidence in ability to do math

40.6517

11.32992

72

72.000

hours of math homework per month

12.1097

2.71471

72

72.000

teacher math support

27.1550

5.62593

72

72.000

confidence in ability to do math

41.4483

10.83725

250

250.000

hours of math homework per month

12.5334

2.95483

250

250.000

teacher math support

27.5500

5.25000

250

250.000

Table D2
Tests of Equality of Group Means
Wilks' Lambda

df1

df2

Sig.

confidence in ability to do math

.980

2.524

247

.082

hours of math homework per month

.977

2.893

247

.057

teacher math support

.981

2.388

247

.094

Table D3
Log Determinants
grouping variable

Rank

Log Determinant

1.00

10.381

2.00

9.389

3.00

10.083

Pooled within-groups

10.011

The ranks and natural logarithms of determinants printed are those of the
group covariance matrices.

Leora Fisher

19

EDPS 612.02 - Psychological Measurement and Statistics


Table D4
Test Results
Box's M
F

18.142
Approx.

1.485

df1

12

df2

269475.410

Sig.

.121

Tests null hypothesis of equal population covariance


matrices.

Table D5
Summary of Canonical Discriminant Functions
Eigenvalues
Function

Eigenvalue

% of Variance

Cumulative %

Canonical Correlation

.043a

80.8

80.8

.203

19.2

100.0

.101

.010

a. First 2 canonical discriminant functions were used in the analysis.

Table D6
Wilks' Lambda
Test of Function(s)

Wilks' Lambda

Chi-square

df

Sig.

1 through 2

.949

12.823

.046

.990

2.500

.287

Table D7
Standardized Canonical Discriminant Function Coefficients
Function
1

confidence in ability to do math

.551

-.133

hours of math homework per month

.412

.772

-.611

.714

teacher math support

Leora Fisher

20

EDPS 612.02 - Psychological Measurement and Statistics


Table D8
Structure Matrix
Function
1
confidence in ability to do math

2
.681*

.239

.655

.700*

-.582

.688*

hours of math homework per month


teacher math support

Pooled within-groups correlations between discriminating variables and


standardized canonical discriminant functions
Variables ordered by absolute size of correlation within function.
*. Largest absolute correlation between each variable and any discriminant
function

Table D9
Functions at Group Centroids
Function
grouping variable

1.00

.265

.043

2.00

-.216

.083

3.00

-.054

-.156

Unstandardized canonical discriminant functions evaluated at


group means

Leora Fisher

21

EDPS 612.02 - Psychological Measurement and Statistics


Part II
What are the strengths and limitations of this approach to research?
What would you do differently if you designed this study?
What can we infer based on the results of your analyses? What will you tell the
researcher?
Results of the previous analyses suggest that the number of hours of math
homework and level of teacher support do influence the level of math confidence in their
ability to do math which support the initial hypothesis. However, the influence of teacher
support is minimal according to my analysis. The strongest support is for hours of math
homework as a positive influence on confidence in math ability, and no significant group
differences were found. The results of this analysis could lead in several directions for
further study such as the number of hours necessary to positively influence math
confidence and/or achievement, the relationship between math achievement and
confidence, and an examination of peer and parental support in addition to teacher
support as influencing math confidence.
One of the key strengths in this analysis was the large sample size, which
provided robustness of results of the MANOVA. The limitations are that no causal
interpretations can be made from the analysis, and possible cofounds exist between
independent variables because of the sheer number of variables provided in the initial
data set. In future studies, it would be advisable to reduce the number of variables for
original data collection. As well, consideration of different groups (SES, male/female,
urban/rural school setting), may provide additional useful data. Finally, a careful
consideration of variables and clearly defined variables in a well-planned study may offer
the opportunity to draw causal conclusions, as supported by specific statistical design
measures.

Leora Fisher

22

EDPS 612.02 - Psychological Measurement and Statistics


References
Stevens, J.P. (2009). Applied Multivariate Statistics for the Social Sciences 5th edition.
New York: Routledge.

You might also like