First: getting started Clear your workspace Anna Petherick: - - PowerPoint PPT Presentation

First: getting started

Clear your workspace Anna Petherick: anna.petherick@politics.ox.ac.uk

Quick revision I

Quick revision II

• Equation for a straight line y=mx+b (or y=b+mx)
• Thus, the equation we estimate to describe the

relationship between x and y takes the form:

• data = model + residuals
• data = intercept + coefficient*IV + error

Quick revision III

Quick revision IV

• A, B, C, D = different

models with same DV

• IVs labelled clearly
• P-values
• Standard errors
• Model fit
• Number of cases.

Model Specification

What to control for?

• Multivariate regression allows us to control for confounders or
• ther possible causes.
• Theory tells us what these things are.
• Two strategies for thinking of confounding variables:
• 1. Think of things that affect y but are not in the regression

model, and then ask yourself whether they might be related to x.

• 2. Think of things that affect x (or things that are related to

and precede x) but are not in the regression model, and then ask yourself whether they might be related to y.

What to control for?

• Absence of control variables can lead to omitted

variables bias

Omitted-Variables Bias

• Bias: expected value of parameter estimate from

sample =/= true population parameter

• Omitted-variables bias: bias resulting from failure

to include a variable that belongs in the model

• So which variables can we omit?
• Can omit Z if completely unrelated to Y
• Can omit Z if completely unrelated to X
• Both of which are unlikely

Omitted-Variables Bias

• Failing to control for relevant variables can lead to

mistaken causal inferences for variables we do include

• Bias may be small, and findings may be right
• Or it may be large, and findings may be wrong
• Consider this when doing own research, and when

reading research articles/books

• Can you think of any other independent variables

that are likely to be related to both x and y?

What NOT to control for?

• Consequences of the treatment variable

X Z Y

• Can lead to post-treatment bias
• Effect of party ID on vote choice
• Do control for race
• Do not control for last-minute voting intentions
• Effect of medicine on health
• Do control for health prior to treatment decision
• Do not control for side effects of treatment

Think Carefully…

• Careful theory: tells us which variables to include

If you want more…

Regression bits, Kellstedt & Whitten, Fundamentals of Pol. Sci..Res. p209-216. Comp Gov:

• 1. Tsebelis, G. and Nardi, D.J. (2014) ‘A Long Constitution is a (Positively) Bad

Constitution: Evidence from OECD Countries’, BJPS. 1-22.

• 2. Ross, M. (2006) ‘Is democracy good for the poor?’, AJPS, 50(4), 860-874.

Challenges the view that democracies have any effect on infant and childhood mortality rates. Pol Soc:

• 1. Jacobsmeier and Lewis (2013). Barking Up the Wrong Tree: Why Bo Didn’t Fetch

Many Votes for Barack Obama in 2012. PS: Political Science & Politics, 46(1), 49-59.

• 2. Inglehart and Norris (2003: Chapter 5)

IR:

• 1. Maoz, Z., and B. Russett. 1993. "Normative and structural causes of democratic

peace, 1946-1986.” American Political Science Review 87: 624–638.

• 2. Gartzke, Erik. 2007. "The capitalist peace." AJPS. 51:166–191.

Next week:

Interactions!

—> Brambor, T., Clark W.R. and Golder, M. (2007) ‘Are African party systems different?’, Electoral Studies, 26(2), 315-323.