marketing analytics: predictive analysis

19
Predictive Analysis Izmir Vodinaj MKMR 310 Fall 2015

Upload: izmir-vodinaj

Post on 12-Apr-2017

209 views

Category:

Marketing


2 download

TRANSCRIPT

Page 1: Marketing Analytics: Predictive analysis

Predictive AnalysisIzmir VodinajMKMR 310Fall 2015

Page 2: Marketing Analytics: Predictive analysis

Project SummaryGoal

Determine the effects of the advertising methods on sales of men and women’s clothing and jewelry.

Data Used (All variables of interest)

1989 -1999 number of catalogs and their pages, phone lines, and men’s and women’s clothing sales.

SPSS Procedures Used

Frequencies, Regression, Select Files, Split File and Graph.

Potential Insights

Determination of continuation of catalog mailing and segmented target content.

Page 3: Marketing Analytics: Predictive analysis

What influences Men’s Clothing Sales?

Number of Catalogs

Number of Pages in Catalogs

Sales of Men’s

Clothing

B1

B2

IV 1

IV 2DV

DV = B0 + B1*IV1 + B2*IV2+ E

H0: None of the independent variables is a significant predictor of men’s clothing sales.

Ha: One or both variables are significant predictors of men’s clothing sales.

Question 1

Page 4: Marketing Analytics: Predictive analysis

Regression Assumptions: Number of Mailed Catalogs and Catalog Pages

Continuous VariablesNormal DistributionLinear RelationshipsNo Multicolinearity No influential cases

Page 5: Marketing Analytics: Predictive analysis

Data FrequenciesQuestion 1

InterpretationLooking at the skewnesses of these variables all being less than 3 and the kurtoses being less than 8, we are able to determine that the variables under study are normally distributed. Furthermore, looking at the individual variables we also see that they are continuous.

Continuous VariablesNormal DistributionLinear RelationshipsNo Multicolinearity

No influential cases

Page 6: Marketing Analytics: Predictive analysis

Checking Linearity

Difference between R Squares is > 0.03 Difference between

R Squares is < 0.03

While number of pages in a catalog and the Men’s clothing sales have a linear relationship, the number of catalogs mailed and men’s clothing sales do not have a linear relationship. This could potentiallybe a sign that the linear model maynot be the best model for this project.

Continuous VariablesNormal DistributionLinear Relationships PartiallyNo Multicolinearity

No influential cases

Question 2

Page 7: Marketing Analytics: Predictive analysis

Checking Multicolinearity

Tolerance > 0.2 VIF < 5

Continuous VariablesNormal DistributionLinear Relationships PartiallyNo Multicolinearity

No influential cases

Due to our tolerance greater Than 0.2 and the variance Inflation factor smaller than 5, We determine that there is noMulticolinearity.

Question 2

Page 8: Marketing Analytics: Predictive analysis

Checking for Influential Cases

Before the Influential Case has been deleted:

I

II

After the Influential Case has been deleted:

Continuous VariablesNormal DistributionLinear Relationships Partially

No Multicolinearity

No influential cases One Deleted

One influential case was found, and it was deleted. After its deletion, the R Square and the Adjusted R Square resulted higher while remaining significant. Beta Coefficients on the other hand changed in different directions. The coef. For number of catalogs increased while the coef. For the number of pages in a catalog decreased.

Question 2

Page 9: Marketing Analytics: Predictive analysis

So, Do The Catalogs Mailed and their Number of Pages Influence Men’s Clothing Sales?

Question 2

Equation with Standardized CoefficientsDV= 0.806*IV1 + 0.119*IV2+ EEquation with Unstandardized CoefficientsDV= - 22690 + 3.38*IV1 + 57.10*IV2+ E

ConclusionOur model can explain around 70% of the variance in men’s clothing sales and the results are significant. The number of catalogs influences the men’s clothing sales more than the number of pages in catalog. For every catalog mailed sales increase by a 0.806 coefficient. Thus we reject the null hypothesis. Finally, we recommend that the store continues and increases the mailing of catalogs.

Number Catalogs Mailed Number of Catalog Pages00.20.40.60.8

1 0.806

0.119

Standardized Beta Compar-ison

Page 10: Marketing Analytics: Predictive analysis

What influences Men’s Clothing Sales?

Number of Catalogs

Number of Pages in Catalogs

Sales of Men’s

Clothing

B1

B2

IV 1

IV 2DV

DV = B0 + B1*IV1 + B2*IV2+ B3*IV3 + E

H0: Adding open phone lines as an independent variable does no change the explanation of variance of men’s clothing sales.

Ha: Adding open phone lines as an independent variable does change the explanation of variance of men’s clothing sales.

Question 3

Number of Pages in Catalogs

B3IV 3

Page 11: Marketing Analytics: Predictive analysis

Regression Assumptions: Phone Lines Added

Continuous VariablesNormal DistributionLinear RelationshipsNo Multicolinearity No influential cases

Question 3

Page 12: Marketing Analytics: Predictive analysis

Data FrequenciesQuestion 3

InterpretationLooking at the skewnesses of these variables all being less than 3 and the kurtoses being less than 8, we are able to determine that the variables under study are normally distributed. Furthermore, looking at the individual variables we also see that they are continuous.

Continuous VariablesNormal DistributionLinear RelationshipsNo Multicolinearity

No influential cases

Page 13: Marketing Analytics: Predictive analysis

Checking Linearity

Difference between R Squares is < 0.03

Difference between R Squares is < 0.03

Adding the phone lines has made it possible that all of the independent variables have linear relationship because the differences between R Squares are under 3%.

Continuous VariablesNormal DistributionLinear RelationshipsNo Multicolinearity

No influential cases

Question 3

Difference between R Squares is < 0.03

Page 14: Marketing Analytics: Predictive analysis

Checking Multicolinearity

Tolerance > 0.2 VIF < 5

Continuous VariablesNormal DistributionLinear RelationshipsNo Multicolinearity

No influential cases

Due to our tolerance greater than 0.2 and the variance inflation factors smaller than 5, we determine that there is nomulticolinearity.

Question 3

Page 15: Marketing Analytics: Predictive analysis

Checking for Influential Cases

Before the potential Influential Case has been deleted:

I

II

After the potential Influential Case has been deleted:

Continuous VariablesNormal DistributionLinear Relationships Partially

No Multicolinearity

No influential cases

No influential cases were found. After suspecting for one influential case and deleting it, the R Square and the Adjusted R Square resulted lower while remaining significant. Beta Coefficients also decreased with the removal of the suspected influential case.

Question 3

Page 16: Marketing Analytics: Predictive analysis

So, does the addition of open phone lines change the results?

Question 3

Equation with Standardized CoefficientsDV= 0.523*IV1 + 0.110*IV2 + 0.431*IV3 + EEquation with Unstandardized CoefficientsDV= - 19041 + 1.95*IV1 + 53.5*IV2+ 322.3*IV3 + E

ConclusionOur model can explain around 78% of the variance in men’s clothing sales, and the results are significant. Adding the phone lines increased the explanation by more than 7%. The number of catalogs still have the highest influence on men’s clothing sales, yet open phone lines have a huge significance as well. For every catalog mailed sales increase by a 0.523 coefficient while for every open phone line sales increase by a 0.431 coefficient. Thus we reject the null hypothesis. Finally, we recommend that the store continues and increases the mailing of catalogs and open new phone lines.

00.30.6 0.523

0.110.431

Standardized Beta Compar-ison

Page 17: Marketing Analytics: Predictive analysis

Comparing the first 5 years to the last 5 years

Number of Catalogs

Number of Pages in Catalogs

Sales of Men’s Clothing

B1

B2

IV 1

IV 2

DV

DV = B0 + B1*IV1 + B2*IV2+ B3*IV3 + E

H0: The explanation of the variance and the significance do not change from the first five years (1989 – 1993) to the last five years (1994 – 1999).

Ha: The explanation of the variance and the significance does change from the first five years (1989 – 1993) to the last five years (1994 – 1999).

Question 3

Number of Pages in Catalogs

B3

IV 3

Year Differenc

esB2

Page 18: Marketing Analytics: Predictive analysis

Question 3

1989 - 1993 1994 - 1999

0.735

0.752

Adjusted R SquareAdjusted R Square

1989 – 1993: St

DV= 0.596*IV1 + 0.005*IV2 + 0.407*IV3

Unstandardized

DV= - 20423 + 2.6*IV1 + 1.8*IV2+ 321.2*IV3 + E

1994 – 1999: St

DV= 0.525*IV1 + 0.162*IV2 + 0.470*IV3 + E

Unstandardized

DV= - 26740 + 1.98*IV1 + 80.5*IV2+ 435.4*IV3 + E

Do the first 5 years differ the last 5 years?

ConclusionThe variance of explanation differs between the years of 1989 – 1993 and 1994 - 1999 with a slightly higher variance of explanation. Number of catalog pages has no significance during the first five years and its beta coefficient increases during the second five years. The open phone lines have a higher beta coefficient during the second five years also, yet the number of catalogs beta coefficient decreased during the second five years. According to these results, we reject our null hypothesis. Finally, it recommended that the managers of the store should increase the number of catalogs but pay close attention at the future trends as its value has been decreasing. In regards to the number of pages and open phone lines, their data suggests increase in value, therefore investing in them moderately would be helpful.

Page 19: Marketing Analytics: Predictive analysis

What Influences the Sales of Women’s Clothing the Most?

Number Cata-logs Mailed

Number of Catalog Pages

Phone Lines Open

Customer Service Reps.

Adverstising-0.2-0.1

00.10.20.30.40.50.6

0.47

0.184

-0.1

0.29 0.26

Standardized Beta Comparison

Stan

dard

ized

Beta

.

InterpretationAfter analyzing five variables and checking for the main regression assumptions (one influencing case was deleted), the number of catalogs mailed has the highest influence on the sales of women’s clothing. The variables are significant, and they explain around 66 % of the variance in women’s clothing sales.

DV= 0.47*IV1 + 0.18*IV2 - 0.1*IV3 + 0.29*IV3 + 0.26*IV3