EMET8002 Case Studies in Applied Economic
Analysis and Econometrics
Semester 1 2025
Computer Lab in Week 5
This week we continue to use the 2015 Programme for International Student Assessment (PISA) dataset for our analysis. PISA assesses skills of 15-year-old students across many countries and involves data on student performance (reading, mathematics and science knowledge), as well as parent, school and teacher questionnaires. The data is publicly available and can be downloaded from the PISA website under the section “SPSS (TM) Data Files (compressed)” (https://www.oecd.org/pisa/data/2015database/). However, these are large files and take some time to download and convert to Stata format. Therefore, we have prepared a smaller merged dataset of the Student and School questionnaire data files which now only includes the relevant variables. You can download this data (“Week4_PISA_data.dta”) from Wattle.
Question 1: Multinomial Logit Regression
We will run multinomial logit models with the following categorical variable as the dependent variable: “org_type” (“What kind of organisation runs your school?”).
(a) Exclude missing values for the following variables: repeated, school_size, age, male, international_lan, mother_edu, father_edu, quiet_study, good_listener, add_math, class_size, computer.
(b) Tabulate the variable “org_type”. How do you interpret the output? Create a label for the variable and values of “org_type” to make interpretation easier and tabulate the same variable again. [Hint 1 : -1 is coded for missing data, 1 means “A church or other religious organisation”, 2 means “Another not-for-profit organisation”, and 3 means “A for-profit organisation”; Hint 2: type in “help label” in Stata for more suggestions].
(c) Run a multinomial logit regression with “org_type” as the dependent variable and the following independent variables: “male”, “age”, “international_lan”, “mother_edu”, “father_edu”. Since the independent variables may contain missing values, include dummy variables that are equal to 1 if there is a missing value and 0 otherwise. [Hint: see “help mlogit” for suggestions]
(d) Why does Stata omit the following two dummy variables: “male_m” and “age_m”?
Run the multinomial logit regression from part c without these two dummy variables.
(e) Calculate the relative risk ratios. How do you interpret the output? [Hint: see “help mlogit” for suggestions]
(f) Calculate predicted probabilities for a student to be at a church school, at a not-for- profit organisation and at a for-profit organisation if they are (i) male and 15 years old, (ii) female and 15 years old, (iii) male and 16 years old, and (iv) female and 16 years old. How do you interpret the output? [Hint: see “help margins” for suggestions]
(g) Compute the marginal effects for the “male” variable for each of the three possible outcomes ofthe dependent variable (church school, not for profit organisation, or for profit organisation). How do you interpret the estimated coefficients? [Hint: see “help margins” for suggestions]
(h) How are predicted probabilities and marginal effects related?
Question 2: Ordered Logit Regression
For this question we use the variable “good_listener” as our dependent variable. This is a categorical variable which ranges between 1 and 4 with 1 being a poor listener and higher values indicating higher listening abilities.
(a) Explain how this dependent variable is different from “org_type” in Question 1 and why we would use an ordered logit regression in this case.
(b) Tabulate the variable “good_listener” and interpret the table. What does the value -1 mean?
(c) Run an ordered logit regression with “good_listener” as the dependent variable and
the following independent variables: “male”, “class_size”, “computer”, “international_lan” and “age”. How do you deal with missing values in the dependent and independent variables? Interpret the output. [Hint: see “help ologit” for suggestions]
(d) Compute odds ratios for the same regression as in part c. How do you interpret the output? [Hint: see “help ologit” for suggestions]
(e) Calculate predicted probabilities for each of the four possible outcomes of “good_listener” for males versus females. Interpret the output.
(f) Do a Brant test to check if the assumption of proportional odds that is required for an ordered logit model holds. Would you still use an ordered logit model? What are the alternatives? [Hint: see “help brant” for suggestions. Most likely, you will need to install the package spost13_ado.pkg, which you can find by typing “findit spost13_ado” in Stata]
Question 3: Preparation for Research Project [not required for problem set]
We strongly recommend that you have chosen a paper by now, as the research proposal is due at the end of week 6, and the data collection process can also take some time. Students are also expected to have started the data application process by now. If you are unsure of the process, ask for help.
(a) Will you need any additional data for your extension? This is optional, as some extensions can be done using the same data as for your replication.
(b) In small groups, discuss what extensions you are planning for your research projects and why they are meaningful, interesting and worth investigating. Give feedback to each other about ways to improve the motivation of your extensions.
(c) Discuss the ethical issues and errors associated with the use of AI-assisted technologies. Note that for this course, the use of AI tools constitutes a breach of academic integrity. (The only exception is when students use Chat GPT or other AI tools when proof-reading, in which case this needs to be disclosed in a statement at the end of your assignment, and you may be asked to provide drafts from before and after the proof-read). If you are unsure of what is appropriate, ask for help.