POLI 30D Midterm 2

Part 1: Identifications

Null hypothesis

the assertion that any apparent difference you see in your sample does not reflect any real difference, but is merely the result of probability

Alternative hypothesis

assertion that there really is some real difference in your sample, over and above whatever is attributable to random fluctuations

Statistical significance

the conclusion that random fluctuations alone can’t account for the size of the effect you observe in your data, so H0 is likely false, and you opt for HA beyond a reasonable doubt

Type 1 error

error occurs when a researcher incorrectly rejects a true null hypothesis. “false positive”

Type 2 error

error occurs when a researcher fails to reject a false null hypothesis. “false negative”

Type 3 error

you arrive at the right answer to the wrong question

Type 4 error

you arrive at the right answer but proceed to interpret it incorrectly

Treatment group

subset of experimental units or participants that receives a specific intervention, manipulation or treatment being tested

Control group

subset of subjects in an experiment that does not receive the active treatment or independent variable being tested

Descriptive statistics

methods used to organize, summarize, and describe the main features of a dataset. Includes the mean, median, mode, range and standard deviation.

Inferential statistics

methods used to analyze sample data in order to make predictions, estimates, or conclusions about a larger population. Includes hypothesis testing, confidence intervals and regression analysis

Dependent t-test

compares the means between two samples, where every observations in one sample affects the choice of what to sample in the other sample group

Independent t-test

compares means between two samples where the selection of one sample does not affect the selection of the other

Restriction

limit participation in the study who are homogeneous with regard to potential confound

Randomization

randomly allocate participants to exposure groups so that the distribution of measured and unmeasured potential confounds should be equal across groups

Double blinding

eliminates bias because both investigators and participants are blinded to whether they are in the treatment or control group

Triangulization

collecting data from various sources

Experimental mortality

subjects leave the experiment non-randomly

Hawthorne effect

behavior by participants is altered because they are aware of their actions being observed

History

besides the treatment, other events take place over time

Instrumentation

measurement itself changes over time

Maturation

steady, long-term trends are also at work

Selection effect

some aspect of selection affects outcome (volunteer bias, only sick receiving medicine)

Spillover

treatment affects control indirectly (people getting vaccinated affects non-vaccinated community members)

Test sensitization

the very act of measuring the cases (when performing a pre-test)

Aggregate fallacy

incorrectly inferring individual behavior from aggregations of individuals

Confounding variables

complicates potential causal relationship as they relate to both X and Y

Simpson’s paradox

when a lurking variable changes the direction of a relationship viewed between two other variables

Chi-square

measuring the association for nominal and ordinal data

Lambda

a measure of how much information about our IV helps us guess our DV

Accord

a type of relationship between two variables in which changes in one variable are related to changes in another variable. An association does not necessarily mean that one variable causes the other

R-squared

value tells us how well our overall model “fits” the data

Independent variable

our “explanatory” variable; the variable we believe causes the dependent variable

Dependent variable

our “outcome” variable; the variable we are trying to explain