GLM

Dept of Security and Crime Science, UCL

---

Tutorial 2, Probability, Statistics and Modelling 2, BSc Security and Crime Science, UCL

---

Outcomes of this tutorial

In this tutorial, you will:

• build and evaluate your own multiple regression models
• build and evaluate your own logistic regression models
• conduct hypothesis testing
  • t-test
  • ANOVA
• perform generalized ANOVA/t-test testing

---

Linear regression

Read the csv file located at data/tutorial2/black_friday/sample_black_friday.csv as a dataframe, and name it blackfriday (Details on the data).

#your code

Familiarise yourself with the dataset:

#your code

Task 1:

Build a linear model that predicts the purchase size based on gender. Use the glm function to do this.

#your code

Task 2:

Assess the model fit using a “fit” metric that tells you how far on average the model mis-estimated the purchase size.

#your code

#your code

Task 3:

Now build two competing models:

• one model with the added variable Marital_Status
• and one model with all terms possible from the two variables Marital_Status and Gender

#your code

Now compare the sum of squared residuals of all three models:

#your code

Now compare the three models statistically. What are your conclusions?

Hint: make sure to check the documentation for the ?anova function. You want to use the “F” test.

#your code

Logistic regression

Load the dataframe called stop_search_met from the file data/tutorial2/stop_and_search/stop_and_search_sample.RData.

#your code

Familiarise yourself with the dataset:

#your code

Task 4:

Subset the data to contain only the two levels of the variable Outcome “Arrest” and “A no further action disposal”.

#your code

Now build a logistic regression model on the Outcome variable as outcome variable modelled through the gender and age of the suspect.

#your code

What are your conclusions about the effects of either predictor variable?

#your code

Task 5:

Expand the model and include the predictor variable Officer.defined.ethnicity.

#your code

How does this affect the model fit?

#your code

Task 6:

Examine the variable Type of the original dataset:

#your code

Now exclude the level Vehicle search and build a logistic regression model on the two remaining levels. Try to answer the question whether race or gender affected the type of stop and search.

#your code

GLM for group comparisons

Use the blackfriday dataset.

Check the documentation in R for the aov(...) function.

Note that an outstanding online help is https://www.rdocumentation.org/.

Task 7:

Does the amount of money spent on purchaes on black friday differ between female and male shoppers?

Show your results with a t-test, an ANOVA and the GLM.

#your code

#your code

#your code

Show that: F = sqrt(t)

Note that for the GLM implementation, you will have to run a model comparison against the model itself using the F test, to obtain the F-value.
Note also that the t.test corrects for unequal variances, which the other two don’t.

#your code

Task 8:

Use an ANOVA and the GLM to test the hypothesis that age affects the purchase size.

#your code

#your code

What is your conclusion?

Task 9:

Bonus question

Use the blackfriday dataset and try to adopt the role of a business analyst. Suppose you are interested in customer profiling and want to target married customers differently than single customers.

Build a model that models the customers’ marital status.

#your code

END

---