✅
💭
🔍
🔸
Labs – Key R Functions to Review:
● attach(data): lets you call variables directly without data$var structure
order(): sorts data
2
●
● aggregate(): summarizes data by group (e.g., means per category)
table(): counts frequencies (e.g., table(regime_type))
1
●
NULL HYPOTHESIS & SCIENTIFIC THINKING
Null Hypothesis
○ Scientists favor the null hypothesis to avoid false positives (Type I
6
errors)
5
○ Null = no effect, no relationship
◆ Ex: “Incumbents do not perform better when the economy is
good.”
Theory vs. Hypothesis
○ Theory = general explanation
4
○ Hypothesis = specific, testable claim derived from a theory
What Makes a Scientific Question?
○ Must be testable, falsifiable, and based on observable evidence
3
◆ Ex: “What are the causes of war?” = scientific
◆ “Does God exist?” = not testable → not scientific
2
What Makes a Good Theory
1
. Causal: Explains how X influences Y (must show a process or
mechanism)
. Not driven from data: don’t allow data alone to drive your theories (like
murder rate in Houston by months)
. Only consider Empirical Evidence: logical theories must be backed up
by rigorous tests; empirical evidence, not ideology, party id, or
metaphysics
. No ideology or normative statements: “we want to get rid of wars /
homelessness / our favorite candidate to win elections” (what are the
causes of wars? What are the reasons behind homelessness? Are
examples of scientific questions).
. Generality: The theory can be applied across multiple cases or settings.
. Parsimonious: Uses the fewest assumptions possible; simple theory.
Also passes rigorous evaluation process and contributes to scientific knowledge.
How to build such a theory? - there is no clear cut formula.
3 Strategies to Build a Theory
. Propose a new Y (with an existing X)
– Rarest and possibly the most creative
. An existing Y and a new X
3
– New independent variable, building new explanations for existing
phenomena (Standing on the Shoulders of Giants)
. Introduce a new Z which modifies an established X —> Y
– X —> Y could be different across different time periods, different
types of individuals, or different geographic contexts.
Does economic voting apply in other democracies?
– No
Research Design
● Definition: A framework for testing theories by examining relationships
between variables using systematic methods.
● Goal: To test a theory against the 4 Causal Hurdles:
. Is there a credible causal mechanism connecting X to Y?
. Can you rule out Y → X (reverse causality)?
. Is there covariation between X and Y? (Do X and Y consistently vary
together?)
. Have you controlled for all confounding variables (Z)?
– Is a third variable (Z) influencing both X and Y?
– Example: More police → less crime? But maybe wealthier areas
have both more police and less crime = confounder
Covariation
● Variables change together
● Necessary for causality but does not imply it
● Different from reverse causality: Reverse means Y → X
● Different from confounding: Confounder (Z) causes both X and Y
Confounding Variable Example:
● X = Rally attendance, Y = Political knowledge
● Z = Education (affects both X and Y)
Problems Measuring Concepts
● Conceptual Clarity: Is the concept clearly defined? E.g., “income” could
mean job income or parental wealth
● Reliability: Produces same result consistently
● Measurement Bias: Systematic error (e.g., always under-reporting
income)
○ Biased but reliable > inconsistent
● Validity:
○ Face Validity: Seems correct on the surface
○ Content Validity: Includes all necessary parts, excludes irrelevant —
most important type
○ Construct Validity: Matches theoretical expectations (e.g., jumping
backwards isn’t an exercise measure)
Variable Types
● Categorical (Nominal): Named groups with no order (e.g., state)
● Ordinal: Ordered, but intervals not meaningful (e.g., agree-disagree)
● Continuous: Infinite values on a numeric scale (e.g., age, GDP)
Statistical Tools
Rank Statistics
● Median, Quartiles, IQR (Q3 − Q1)
● Used when data is skewed or has outliers
Central Limit Theorem (CLT)
● Sampling distribution of the mean becomes normal (bell-shaped) with
large samples
● Justifies statistical inference even if population is skewed
Statistical Inference
● Making claims about a population from a sample
● Accepts uncertainty — “best guess” from evidence
Bivariate Tests
IV Type DV Type Test Type
Categorical Categorical Chi-Square / Tabular
Continuous Categorical Logit / Probit
Categorical Continuous T-Test / Difference of
Means
Continuous Continuous Regression /
Correlation
Regression & Multicollinearity
What is Regression?
● Statistical method for estimating relationships
● OLS = Ordinary Least Squares
● Visualized as a line minimizing prediction error
Key Terms:
● Independent Variable (IV) = Predictor
● Dependent Variable (DV) = Outcome
● Intercept (α) = Value of Y when X = 0
● Coefficient (β) = Effect of one unit increase in X on Y
● P-Value: Probability result is due to chance (significant if < 0.05)
● Slope = Coefficient → rise/run of the line
Dummy Variable
● Encodes binary info (e.g., male = 0, female = 1)
● Allows categorical data in regression
Multicollinearity
● Occurs in multiple regression (not bivariate)
● Two IVs are highly correlated → hard to separate effects
● Use VIF (Variance Inflation Factor):
○ VIF > 10 = multicollinearity problem
Example: Dem = 0.2 + 0.24 * GDP - 0.32 * War
● Intercept = baseline value when predictors = 0
● GDP coefficient = positive relationship with democracy
● War coefficient = negative relationship
● Interpret signs, sizes, and p-values