Panel Surveys

 

POLSCI 4SS3
Winter 2023

Announcements

  • Lab 2 deadline extended to Friday, January 27 (no new lab today)

  • Sign up for groups!

  • By 9 AM tomorrow or I will put you wherever

Last week

  • Overview of MIDA approach to research design as programming

  • Representative surveys as the gold standard of public opinion research

  • Challenging to decide:

    • What to ask
    • Who to ask

  • Lab: Get to know R, practice simulating a survey with random sampling

Today

  • Start thinking about cause and effect

  • Panel surveys: Survey the same sample multiple times

  • Lab: Revisit last week’s lab

Cause and Effect

Elements of a model

  1. Signature

  2. Functional relations

  3. Probability distribution over exogenous variables

Part 1: Signature

  • : Describes variables and their ranges

  • Two kinds of variables

  • Endogenous: Generated from within the model

  • Exogenous: Generated from without the model

Types of exogenous variables

  1. Anything explicitly (or assumed as) randomized

    • Mostly experimental treatment assignment

    • Denoted by \(Z\)

  2. Anything unobserved by the model

    • Otherwise we would be in trouble!

    • Denoted by \(U\)

Types of endogenous variables

  • Anything else

  • Outcomes: The things we ultimately want to understand (\(Y\))

  • Moderators: Variables that modify effects (\(X\))

  • Mediators: How or why something has an effect (\(X\))

  • Confounders: Introduce non-causal dependence (\(X\))

Part 2: Functional relations

  • : Set of functions that produce endogenous variables

  • Two ways to express functional relations

  1. Structural causal models (today)

  2. Potential outcomes framework (next week)

Structural causal models

  • Use Directed Acyclic Graphs (DAGs)

  • Directed: Connected by arrows

  • Acyclic: Not cyclical, usually end in outcomes

  • Graphs: Visual representation as nodes and edges

  • They represent nonparametric causal models

Example

DAG for \(Y = f_y(Z,U)\)

Part 3: Probability distribution over exogenous variables

  • An explanation of how exogenous variables are generated

Examples

  • \(Z \sim \text{Bern}(p)\) with \(p = 0.1\)
  • \(U \sim N(\mu, \sigma)\) with \(\mu = 0\) and \(\sigma = 1\)

Panel Surveys

What are panel surveys?

  • Surveys where the same participants are asked questions at multiple points in time

  • Usually measure outcomes at every time (but not necessary)

  • More common among convenience samples (e.g. students, twins)

  • The name comes from their data structure

Panel data

Balanced panel
ID year income age sex
1 2016 1300 27 M
1 2017 1600 28 M
1 2018 2000 29 M
2 2016 2000 38 F
2 2017 2300 39 F
2 2018 2400 40 F

Panel data

Unbalanced panel
ID year income age sex
1 2016 1600 23 M
1 2017 1500 24 M
2 2016 1900 41 F
2 2017 2000 42 F
2 2018 2100 43 F
3 2017 3300 34 M

What are panel surveys for?

  1. To measure attitudes in a population over time
  1. To understand the effect of events occurring between waves

Challenge

  • Panel attrition: Participants may drop out from follow up waves

  • It may offset the benefit of conducting a panel survey

  • It may depend on factors relevant to the study

Example 1

Example 2

Next Week

Sensitive Questions

Focus on: Which research design seems more appropriate to elicit honest answers?

Break time!

 

Lab