Psych stats exam 2: Hypothesis Testing & Experimental Design! (generally)

alpha (significance level)

establishes the value of p that, when observed, determines when the null hypothesis should be rejected

-denotes the chance of rejecting the null hypothesis when it is actually true

-"the results were significant at the .05 alpha level"

p-value approach

calculates exact probability of the test statistic and compares that to alpha

- p-value > alpha: fail to reject the null

- p-value < alpha: reject the null

directional/one-tailed hypothesis

specifies whether the sample mean will be less/greater than the population mean

-critical region falls on one end of the distribution

-easier to reject null because alpha is not split

-one tailed tests can miss unexpected findings

-ex: "a happier dog will eat more"

non-directional/two-tailed hypothesis

predicts sample mean will be different than population mean, but does not state a direction of the difference

-critical region falls on both ends of the distribution, alpha/2...makes significance more difficult to find

-better for basic science; can make interesting findings in both directions

-ex: "mood will affect the appetite of a dog"

assumptions of the one sample Z-test

1. sample must be drawn for a population with a normal distribution, or have sufficiently large sample

2. standard deviation of the population is known

-critical for SE calculation

-without it, cannot determine variability, and therefore cannot determine if sample does or does not come from target population

3. dependent variable is ratio or interval

p-value

tells us the probability of our data under the assumption that the null hypothesis is true

what does the p-value NOT tell us

does NOT tell us the probability of the hypothesis being true

left tail test

alternative is less than

-H0>

-HA<

-"CBD reduces anxiety"

right tail test

alternative is greater than

-H0<

-HA>

-"Chronic gamblers' DA activation is greater than non-gamblers"

One Sample Z-test

comparing one sample to the population

-not used very often because we very rarely know the population's standard deviations

p-values approach

-finding the exact p-value for test statistic, and determining what to do about the null based on p vs. alpha

One sample t-test

used when population standard deviation is unknown

-same procedure as one sample z-test, except using t-table

-must use critical values approach (from n-1 df), not p-value approach

type I error

null hypothesis is true, but we mistakenly reject it

-false positive

-determined by alpha value:

alpha .05 means we make type I error 5/100 times/5% of the time

type II error

null hypothesis is false, but we fail to reject it

-false negative

-probability of this error is beta (β)

-increasing sample size decreases standard error, which increases t-score; increases β means increased power

requirements of designing an experiment

1. manipulating the IV; must be at least 2 conditions (control and treatment)

2. random assignment

-each participant must have equal chance of being assigned to any condition

-eliminates chance that there's any systematic differences between groups

two types of manipulations

1. between subjects

2. within subjects

between subjects manipulation

2 groups: 1 in control, 1 in treatment

-mean of control group is way of estimating mean of population

-requires larger overall sample size

-limits participant effects as participants don't know what's being manipulated

within subjects manipulation

give all participants the same treatment, measure the before vs. after (same participants, measured twice)

-pre-test measurement is best estimate for population mean

-helps reduce potential error from random sampling

-could be more reliable; bigger sample size (which may make it easier to reject the null)

-prone to participant effects as manipulation can be obvious

-aka repeated measures design

Random assignment improves...

internal validity

-our ability to conclude that the observed change in the dependent variable was a result of our experimental manipulation of the independent variable

Random sampling improves...

external validity

-our ability to conclude that the findings of our study will generalize to the wider population (rather than just the selected group of participants we sampled)

true experiment

random assignment to experimentally manipulated conditions

-better-allows us to make causal claims

quasi experiment

differences between participants in the study conditions are not experimentally manipulated

-involves individual differences/subject variables

-does not allow for causal claims because there is no manipulation

strengths of between-subjects designs

limits participant effects

-different participants serve as control vs. experimental manipulation, so there isn't a chance for participants to learn (e.g. practice effects) or for the order of the tasks to matter (e.g. order effects)

weaknesses of between-subjects designs

require recruitment of a larger sample

-need different participants to assign to experimental condition and to control condition

strengths of within-subjects designs

easier to reject null hypothesis (more power) with a smaller sample

-participants serve as their own control group, so you don't have to recruit a separate group of participants, but still have the same power to detect effects

weaknesses of within-subjects designs

participant effects

-order effects

-practice effects -demand characteristics