Last part

Question 1

logistic regression

Answer 1

used mainly when DV is NOT A # – used to predict categories

Question 2

basic types of logistic regression:

Answer 2

logistic or probabilistic: binary outcomes

multinomial: discrete outcomes, not in relation to each other (categorical)

ordered: discrete oucomes on a scale in relation to each other
- poor, fair, excellent

not logistic: count-based outomes, even though they are whole numbers

Question 3

discrete variable

Answer 3

A variable that can take on specific, separate values — usually whole numbers — and nothing in between.

ex: # of children, # of protests

used in: OLS, logistic, ordered logistic

Question 4

continuous variable

Answer 4

can take on any #

ex: voter turnout %

Question 5

logistic or probabilistic regression

Answer 5

2 options (yes/no)
OLS reg does not work with this bc distribution is entirely 0 to 1

Question 6

logistic v. linear regression

Answer 6

logistic = binary DV
- OUTPUT: probability or log-odds
- Use logistic when outcome is yes/no OR Categories
- -Change in coefficients: change in log odds/odds ratio

linear = continuous DV
- OUTPUT: raw value
- use logis
- probability #
- use linear when outcome is a #
- -Change in coefficients: change in DV
- Probability between 0 and 1

Question 7

binary outcomes

Answer 7

any DV that takes on two outcomes

Distribution curve is 0 or 1, so equation for OLS regerssion doesnt work

output can be read the exact same as a linear regression

Question 8

ORDINARY LEAST SQUARES

Answer 8

Most common linear regression, used when DV is continuous

uses normal basic equation of a regression

OLS finds the best-fitting line through the data by minimizing the squared distance between the actual values and the predicted values.

That’s why it’s called least squares — it minimizes the sum of squared errors!

Question 9

use of logarithms

Answer 9

logarithm = inverse of an exponent

help us assess curves like we see in binary data

taken using exponents. the base with an exponent of x equals x

Question 10

natural logarithm

Answer 10

NATURAL LOG is base “e” (an irrational derivative) bc elnx = x

e^{\ln x}=x

It’s log base e, where e ≈ 2.71828.

It’s called “natural” because it arises naturally in calculus, growth processes, and probability models.

**Natural logarithms are used to linearize exponential relationships, model log-odds, handle skewed data, and measure growth rates — they’re essential in regression and probability modeling.
**

Question 11

why is the exponent negative? why is it sometimes written as a positive

Answer 11

negative exponent constrains value between 0 and 1

Question 12

why isnt the y intercept constrained to 0 and 1

Answer 12

bc coefficients in logistic models are read kind of differently

Question 13

IN A REGRESSION TABLE: INTERCEPT

Answer 13

Log-odds of segregation when all Xs = 0 (not directly interpretable)

Question 14

IN A REGRESSION TABLE: DV

Answer 14

First thing after formula = (dv)

Question 15

negative exponents

Answer 15

use these to constrain output between 0 and 1 (valid probabiliites)

Question 16

y-intercept

Answer 16

NOTE: Linear y-intercept = value of Y when all Xs are 0. in Logistic, remember we are predicting log-odds, NOT raw numbers.

y intercept is NEGATIVE of location parameter divided by rate parameter
location parameter / rate parameter

(location parameter = point at which the porbability = 0.5 //midpoint basically)

(rate parameter = tells you how fast something happens)

Question 17

why is y-intercept between 0 and 1

Answer 17

bc the regression is in log-odds, not raw probabilities

Question 18

odds ratios

Answer 18

ratio of an outcome relative to its alternative

AKA likelihood of y = 1 divided by likelihood of y = 0

Question 19

when reading coefficients for a logistic regression…

Answer 19

coefficients are listed as odds ratios

read in relation to 1 NOT 0 – so lower than 1 is negative, higher than 1 is positive

Question 20

IYENGAR and WESTWOOD (2014)

Answer 20

differentiates between policy, identity and affective polarization

identity polarization: alignment with others based on party affiliation, not policy

affective poliarzation: hostility towards other members of other political parties

Question 21

WHEN TO USE LOGISTIC REGRESSION

Answer 21

ALL ABOUT DISTRIBUTION!
- binary distributions dont follow OLS

!!!!Standard errors and Test Statistics work the same way – distance to line!!!!

Question 22

what to do about categorical and ordinal variables

Answer 22

need to adjust the logistic regression….called multinomial and ordered logistic regression

Question 23

categorical DV

Answer 23

Takes on several discrete, non-ordered categories

outcomes not assessed in relation to each other

in formula, K is the baseline outcome

Question 24

ordinal DV

Answer 24

each category is ordered, so PROBABILITIES are SUCCESSIVE

each outcome hsas to be compared in relation to other outcomes.

on a 5-point LIKERT scale (very poor, poor, fair, good, very good)

Question 25

V-DEM index

Answer 25

varieties democracy. five components of democracy w/ subcomponents. resulting index is normalized from 0 to 1

Question 26

inter-coder reliability

Answer 26

typically 4+ ppl collecting the same data and **cross-referencing results**. Teams of coders include project managers, research assistants, and experts.

Question 27

Compiling data

Answer 27

Standardized scales using aggregate probability distributions across coders Alternate versions using average, median, ordinal scores, and high-low estimates These are “confidence intervals” for probability densities Each variable has around 10 versions available in the codebook

Question 28

v-dem index

Answer 28

varieties democracy five components of democracy w/ subcomponents resulting index is normalized from 0 to 1

Question 29

inter-coder reliability

Answer 29

typically 4+ ppl collecting the same data and cross-referncing results. Teams of coders include project managers, research assistants, and experts.

Question 30

objective vs subjective criteria

Answer 30

objective: Facts based on real things like laws, stats, and official records. subjective: Used when things aren’t clear-cut, like judging if one party controls vote counting.

Question 31

vignette coding

Answer 31

a method where experts rate complex or subjective political phenomena (like attacks on the judiciary) using tools like Likert scales (from 1-5), rather than relying on hard data. **It helps measure concepts that can't be captured with purely objective indicators.**

Question 32

LIKERT SCALES

Answer 32

A common survey tool that asks respondents to rate their **agreement** or **perception** on a **scale** (e.g., from “Strongly Disagree” to “Strongly Agree”), usually 5 or 7 points, used to quantify attitudes or assessments. Use in cross-sectional data requires more individual judgment from researcher.

Question 33

CROSS SECTIONAL DESIGN

Answer 33

Snapshot in time; no time variable. Best for variables that don’t change much over time (e.g., constitution). Useful for multilevel models (grouping individual data by country).

Question 34

time series design

Answer 34

data that changes over time (e.g., regime change). Easier at macro-level (countries), hard at individual-level. Requires Fixed Effects to control for units (e.g., year, country).

Question 35

FIXED EFFECTS

Answer 35

Control for unobserved differences across groups (like countries or years). Fits separate regression lines within each group. Used only when you have data for the whole population.

Question 36

SCALING

Answer 36

transforming, adjusting, or interpreting data so that it can be meaningfully compared or applied across different levels, units, or contexts. easiest way of scaling data is to set values between 0 and 1 in this case, a value corresponds to a percent of the max value but data can also be normalized or transformed (set to a particular curve) transforming usually means using natural logs to account for skew

Question 37

NORMALIZING

Answer 37

Normalizing data means data is fit to a normal distribution normalized data could be scaled, corresponding to percentile in normal distribution - Could also be a raw # of SDs away from mean raw value of V-DEM data are usually normalized as STANDARD DEVIATIONS

Question 38

unit of analysis

Answer 38

under what **parameters** are you collecting data? country-level data is exhaustive -- bc pulls from whole population can also do this with any JURISDICTION - a district, province etc individual-level data is not -pulled from a sample

Question 39

cross-sectional design

Answer 39

snapshot in time comparing macro-level effects w/out the element of time country-level variables that dont change much over time, like constitutional scope

Question 40

simpsons paradox

Answer 40

It’s when a trend appears in aggregate data, but **disappears** or **reverses** when you break it into subgroups. happens when u mix 2 diff levels of analysis only works if you have complete data of the target population. You might find a strong relationship at the country level (like: "More immigration restrictions → fewer women in parliament") But that trend could vanish or reverse when looking at specific regions, income levels, or years.

Question 41

FIXED, RANDOM AND MIXED EFFECTS

Answer 41

FIXED: deployed when u have whole population RANDOM: unit controls when you have a partial population MIXED: fixed and random effects together for multilevel data

Question 42

CROSS-SECTIONAL V. TIMES SERIES V. FIXED EFFECTS

Answer 42

cross sectional = snapshot in time times series = observes changes over time fixed effects = controls for group level differences

Question 43

count models

Answer 43

count model = type of statistical model used when your dependent variable is a count of events -- like # of protests, # of laws passed or # of times someone voted negative binomial more often used bc more robust compared to POISSON

Question 44

origins of poisson distribution

Answer 44

examines distribution of occurences - the rate at which diff outcomes occur

Question 45

how does negative binomial build on the poisson distribution

Answer 45

🔁** The Poisson Distribution (Baseline)** Used for count data (e.g. number of protests, number of asylum claims). Assumes that the **mean = variance** — this is called *equidispersion*. Good for when your data is neatly distributed with little variance. ❗ The Problem: **Overdispersion In real life, the variance is often greater than the mean** (i.e. overdispersion). Poisson struggles here — it underestimates standard errors and can inflate significance. ✅ The Negative Binomial: A Fix The Negative Binomial distribution generalizes Poisson by introducing an extra parameter to model overdispersion. Here’s how it builds on Poisson:gauges rate of failures given # of trials

Question 46

LAMBDA

Answer 46

rate parameter It tells you the average number of events per interval if lambda = 1 means you expect 1 event per interval

Question 47

NEGATIVE BINOMIAL

Answer 47

Regression show incidence Rate Ratios -- outcomes divided by non-outcomes + it accounts fo overdispersion The Negative Binomial distribution becomes more symmetric and “normal-looking” as r increases When r is small (like r = 1), it looks very skewed, almost like an exponential or geometric distribution