Video 1: basic statistical analysis
Screening the dataset:
1. No missing values
2. Logical values?
3. Reverse coding
If the data is ready:
1. Describe and summarize data → descriptive analysis (frequencies, percentage, mean std
dev,
Inferential analysis:
- Hypotheses testing
- Compare means → one sample T-test
- T-value = (mean sample-mean test value)/ std.error mean → must be >2 (critical value)
Differential analysis
- Hypotheses testin
- Compare means
- Independent Samples > 2 → one way ANOVA
1. Related sample → comparing response of the same individual, amongst each other
→ paired sample T-test
2. Chi-square test → comparing the means of non-metric variables (nominal, bv age) if
P<0.05, reject H0 → the difference between the means is significant
Associative analysis
- Checking correlations between variables
- Magnitude of the correlation coefficient (R ), magnitude, and significance
- Pearson’s correlation → negative: the greater the household size, the lower the evaluation
- Pearson’s correlation → between 0.1-0.3 = weak, 0.3-0.5 =medium, >0.5=strong
Video 2: Scale Analysis case
Factor analysis:
- KMO’s and Barletts test → whether it is appropriate to conduct factor analysis
(intercorrelations between variables)
→ KMO > 0.5
→ significance > 0.05
- Total variance explained → Look at EigenValues > 1 → this is the amount of factors (max
of 5 factors)
- How much of the variance is captured by the factors loading
- Communalities table: >0.3, lower amount of factors, lower number for communalities
- Rotated component matrix: >0.5, must load on high on 1 factor and low on the others
→ if item does not group together with other items on same factor, or <0.5 → exclude
item
Task 2: Reliability assessment
, - Group the items that load on the factor together
- Cronbachs alpha > 0.7
- Look at Cronbachs alpha if item deleted → if it doesn’t go up, accept the cronbachs alpha
as it is
- If alpha > 0.7, but it goes up if item deleted → don’t delete that item → the more items
you give up, the more the validity will go down
Task 3: Compute mean scores on dimensions (summated scales)
- Compute the mean scores of the items that are grouped together
- You can do for example regression with these new variables → for multicollinearity
Video 3: Market Response model
To measure the impact of different marketing inputs on consumer demand
→ test whether the marketing actions had any effect on sales
→ evaluate the marketing actions
→ what-if to develop future market decisions
1. Analyze trends and relationships among variables
→ trends: plot per variable
→ relationships among variables: scatterplot
2. Run regression analysis using different models
- Model fit: Adjusted R squared → the model explains (r2) of the variance of the DV
- Coefficient table: Sig <0.05
→ B: For every unit increase in price, the sales increases by (b value) holding everything
else constant
- Model fit opt 2: Sales vs predicted sales
→ Predicted sales = a + b1price + b2Ad + error
→ A= constant, b1 = b value price, b2= b value Ad, error =unknown
→ the value of price and ad are given
→Residual Error is the difference between actual and predicted sales
- Weak fit of model → create new model → cummulated advertising spend
→ regression with CummAd as new IV
- Collinearity statistics:
→ Tolerance > 0.1
→ VIF < 10
- Which IV has a greater impact on the DV → standardized coefficient Beta → the
highest value has the largest impact on the DV (look at absolute values)
- When you find the best model fit → what -if simulations to guide future marketing
decisions
→ what happens if you stop advertising after one year?
→ what happens if you increase advertising budget with 50%?
Screening the dataset:
1. No missing values
2. Logical values?
3. Reverse coding
If the data is ready:
1. Describe and summarize data → descriptive analysis (frequencies, percentage, mean std
dev,
Inferential analysis:
- Hypotheses testing
- Compare means → one sample T-test
- T-value = (mean sample-mean test value)/ std.error mean → must be >2 (critical value)
Differential analysis
- Hypotheses testin
- Compare means
- Independent Samples > 2 → one way ANOVA
1. Related sample → comparing response of the same individual, amongst each other
→ paired sample T-test
2. Chi-square test → comparing the means of non-metric variables (nominal, bv age) if
P<0.05, reject H0 → the difference between the means is significant
Associative analysis
- Checking correlations between variables
- Magnitude of the correlation coefficient (R ), magnitude, and significance
- Pearson’s correlation → negative: the greater the household size, the lower the evaluation
- Pearson’s correlation → between 0.1-0.3 = weak, 0.3-0.5 =medium, >0.5=strong
Video 2: Scale Analysis case
Factor analysis:
- KMO’s and Barletts test → whether it is appropriate to conduct factor analysis
(intercorrelations between variables)
→ KMO > 0.5
→ significance > 0.05
- Total variance explained → Look at EigenValues > 1 → this is the amount of factors (max
of 5 factors)
- How much of the variance is captured by the factors loading
- Communalities table: >0.3, lower amount of factors, lower number for communalities
- Rotated component matrix: >0.5, must load on high on 1 factor and low on the others
→ if item does not group together with other items on same factor, or <0.5 → exclude
item
Task 2: Reliability assessment
, - Group the items that load on the factor together
- Cronbachs alpha > 0.7
- Look at Cronbachs alpha if item deleted → if it doesn’t go up, accept the cronbachs alpha
as it is
- If alpha > 0.7, but it goes up if item deleted → don’t delete that item → the more items
you give up, the more the validity will go down
Task 3: Compute mean scores on dimensions (summated scales)
- Compute the mean scores of the items that are grouped together
- You can do for example regression with these new variables → for multicollinearity
Video 3: Market Response model
To measure the impact of different marketing inputs on consumer demand
→ test whether the marketing actions had any effect on sales
→ evaluate the marketing actions
→ what-if to develop future market decisions
1. Analyze trends and relationships among variables
→ trends: plot per variable
→ relationships among variables: scatterplot
2. Run regression analysis using different models
- Model fit: Adjusted R squared → the model explains (r2) of the variance of the DV
- Coefficient table: Sig <0.05
→ B: For every unit increase in price, the sales increases by (b value) holding everything
else constant
- Model fit opt 2: Sales vs predicted sales
→ Predicted sales = a + b1price + b2Ad + error
→ A= constant, b1 = b value price, b2= b value Ad, error =unknown
→ the value of price and ad are given
→Residual Error is the difference between actual and predicted sales
- Weak fit of model → create new model → cummulated advertising spend
→ regression with CummAd as new IV
- Collinearity statistics:
→ Tolerance > 0.1
→ VIF < 10
- Which IV has a greater impact on the DV → standardized coefficient Beta → the
highest value has the largest impact on the DV (look at absolute values)
- When you find the best model fit → what -if simulations to guide future marketing
decisions
→ what happens if you stop advertising after one year?
→ what happens if you increase advertising budget with 50%?
