inta 2010 emperical methods midterm

emperical evidence

measured systematic trends

anectdotal evidence

personal experience

building theories

not data, theory, data, confirm/deny theory

a theory contains

answer and why

a hypothesis contains

guess statement, no why

powners 4 parts of theory

expectation, causal mechanism, assumptions, scope conditions

in P. 4 parts of theory, what is the expectation

what answer does the theory provide to the question

in P. 4 parts of theory, what is a causal mech

why causes x to y, why?

in P. 4 parts of theory, what is assumptions

what has to be true for the theory to hold

outcome we want to explain

dependent variable

what we measure, what is the explanatory concept

independent variable

what are K&W 4 hurdles for causality

credible causal mechanism, could y lead to x, covariation between x&y, are confounding vars being controlled

in k&w hurdles, what are the options for covariation

do the variables move together in a positive relationship, or apart in a negative relationship

cross sectional variation

change over space

longitudinal variation

change over time

hypothesis contains what

null hypothesis

what is a null hypothesis

expect no change or no patterns

causality

is there a causal relationship and whats the substantive effect

what do we worry about for causality

biased causal effects

what are the 3 biases for causality

simultaneity bias, common cause bias, selection bias

what is simultaneity bias

comes from reverse causality, x to y/y to x

what is common cause bias

comes from confounding variables where z makes it seem like x and y have a relationship

what is selection bias

comes from selection effects; only looking at certain values of DV or not being thorough with causal relationship

counterfactual theory of causation

were x to be different than y would also be different

fundamental problem of causal inference

cannot assign the same unit to both treatment and control at the same time

iv should be exogenous or endogenous

exogenous

what does exogenous mean

no reverse causality, doesn't depend on other things that effect DV

what are control variables

part of IV, potential confounding vars

3 measurement metrics

categorical, ordinal, continuous

what are categorical vars

no universal ranking; ex) is a country democratic

what are ordinal vars

has rankings but not equal unit diffs; ex) how democratic is a country

what are continuous vars

equal unit diff and universally held rankings; ex) income levels

reliable

consistent when repeated

measurement bias

systematic under/over reporting of values for var

three types of validity

face, content, construct

what is face validity

does it make sense?

what is content validity

is the measure complete?

what is construct validity

is it reasonable given measures of related concepts

steps of operationalization

1) conceptual clarity
2) measurement metric
3) reliability
4) measurement bias
5) validity

what are the types of research design

experiments, small N, large N

what are the types of experiments

randomized control trials, natural

tools to identify causality

randomization, control

why is causality important

extract implicit biases, change outcomes of interest, identify policy impacts

what are the two types of research designs

experiments and observational

basic steps of experiments

sample subjects, randomly divide subjects into groups, measure and compare values of DV between groups

diff types of sampling

convenience, random, representative

what is the gold standard of research

experiments

2 problems when establishing causality for experiments

confounding factors, endogeneity

necessary assumptions for experiments

randomization, excludability, non interference

pros for experiments

highly transparent, replicable, allows for tests of statistical significance

cons for experiments

ethics, external validity, not all variables manipulable

what is an observational design

design where researches does not control administration of treatment

steps for observational design

decide on type of study, gather data on IV, DV, and controls; model relationships between variables

steps for deciding on type of study for observational design

type of variation, units of observation, type of data

steps for gathering data for observational design

population and sampling technique, primary or secondary data, variables

what is the most similar (method of difference)

for small N studies, look at diff dv, diff iv, and same controls; then IV must be cause of DV

what is most different (method of direct agreement)

for small N studies, look at same DV, same IV, different controls then one IV same must cause DV

pros to small N studies

answers how; high plausibility; eliminate endogeneity; high internal validity

cons to small N studies

external validity, selection effects; no randomization; ethical concerns

large N observational designs

collect as much data as possible and use stats to identify patterns;

pros to large N design

feasible, cheap, external validity

cons to large N design

low internal validity, hard to identify causality;

what is variation ratio

V = 1 - (number of modal cases) / (total number of cases)

Shows percentage of cases outside modal category

what is variance

measure of dispersion of variable around mean; SD^2

what descriptive statistics needed for categorical

mode, var ratio

what descriptive statistics needed for ordinal

mean, median, mode, variation ratio, and IQR

what descriptive statistics needed for continuous

mean, median, mode, range, IQR, variance, standard deviation

gyst of CLT

keep taking samples, getting means, and plotting them, the means would eventually form a normal distribution with a mean = true population mean and standard error = standard deviation of population

equation for SEx

SEx = SDx/sqrt(n)

what gets us from what we have to what we want

CLT

confidence intervals definition

gives set range of where we think true population value is likey to be located

confidence interval

1-(fish figure)

margin of error =

t(crit) * SEx

if confidence % is outside MEx

capturing systematic pattern

if confidence % is statistical tie

capturing random pattern

type 1 error

we say smth occured but smth did not occur

type 2 error

we say smth did not occur but smth did occur

if tcrit is less than tx

reject our null

if tcris is greater than tx

fail to reject our null