Module 1 + 2 Flashcards
analyzing data involves being part (careers)
- good detective
- honest lawyer
- good storyteller
what test is used to determine if scores for a sample of ppl are different from a theoretically specified score
single sample t test
what test is used to determine if scores for a sample of people are different at two points in time
two sample t test
in comparative stats there are _____ or ____ explanations for claims
- systemic, chance
- ex random chance, systemic influence
NHST
- Null hypothesis statistical test
- to determine if the observed difference is different than if it were due to chance
- dominate procedure for differentiating chance and systemic influence
if you test your hypothesis and determine that chance is ruled out do you ;
a) accept that change is due to systemic reasons only
b) accept that change is due to a combo of chance and systemic reasons
b)
ablesons magic criteria
- properties of data, analysis, and presentation that determine strength of research claim
- Magnitude
- articulation
- generality
- interestingness
- credibility
random error often evens out in ____ sample sizes
larger
the smaller the sample size, the ____ the difference needs to be in order to be significant
larger
alpha (0.05 in psyc) can also be referred to as
tolerable difference
independent samples t test
- two samples that are independent from one other / drawn from separate populations that are then compared to determine if there is a difference
formula for independent samples t test in words
t = sample data - hypothesized population perameter/estimated standard error
what do larger t values indicate
- greater likelihood of difference from hypothesized value
- two scored differentiate from one another
formula for independent samples t test in symbols
t = (x1-x2)-(μ1-μ2) / SE
or
t= x1-x2/SE
x1/2=means from samples
μ=means from populations
in the null hypothesis, μ1-μ2= _____
0, there is no difference between populations
standard deviation
- how far your sample is dispersed from the mean
- spread around the average score
Standard error formula
Sx= S/√n
standard error formula for an independent t test
Sx1-x2=√ (s^2/n1 + s^2/n2)
degree of freedom formula
df= (n1-1) + (n2-1)
how to use df to calculate a missing number if you have 2/3 numbers and the mean
- there is only one # that can work
- df= unique answer/#
- any property from the sample can be used to determine other value
property of sample formula
sample mean (n-1)
n
sample size
t or f: as df increases, values tend to be spread further from 0
false, they cluster closer to 0 (above 120=df, data is very close)
alpha
- probability of messing up that is acceptable
- 5% (0.05)
- 2.5% in each tail of two tail test and 5% in one side in one tailed test
because test gives no direction it is called a _______ test, with 2.5% representing the most extreme ____ and ____ values
two tailed, positive and negative
type 1 error
- rejecting the null when its true
- finding a difference when there is no difference
type 2 error
- failing to reject the null hypothesis
- finding no difference when there actually is one
replication crisis
- ppl doubt psych research because when studies are replicated, different results occur/lower rates occur
One tailed tests
- aka directional tests
- only considering extreme t values in one direction (ie positive)
- rather than 5% in both tails, 5% in one side
- more wiggle room
- p value is half that of a two tailed test
lopsided test
- compromise between one and two tailed tests when researcher has directional prediction
- weight the tails of distribution (more liberal for predicted direction and conventional for unexpected direction)
what is the widely accepted standard for lopsided tests
- there is none
- any as long as you can defend/justify
conventional level for type l and type ll errors
- type l: 0.05
- type ll: 0.2
power
- 80% is goal/standard
- probability that stat test will correctly reject a false null hypothesis
- oppositely related to type ll error
- power=1-β
determinants of power
- alpha level (stricter=lower power, under researcher control)
- sample size (larger=bigger power/lower SE, under researcher control)
- magnitude of effect/effect size (larger IV effect=bigger power, somewhat under researcher control)
how to calculate power of test
- stat tables/programs calculate using: alpha, sample size, and magnitude of effect
how can power help with sample size planning
- before a study, can help determine appropriate sample size
- specify alpha (0.05) and desired power (0.80), make assumption of magnitude then you can calculate sample size needed to get all the values
assumptions about independent samples t tests
- independence of observations
- normal distribution for each group
- equality of variance in outcome variable across groups
repeated measures t test
- testing diff between two means for same sample of ppl
- usually longitudinal w/ intervention in between
- same outcome under different conditions
- aka paired samples t test
Sample data difference scores are rep’ed by _____ in equations whereas the mean of different scores is ____
D, D (with line above)
formula for repeated measures t test
t= (D_-µD)/S(D_)
or
t=D_/S(D_)
µD=mean of difference scores in population
SD_=standard error of sample mean of difference scores
in repeated measures t tests, SD_ formula
SD_= S/√n
S=standard deviation for difference scores
df in a repeated measures t test formula
df= n-1
(because there is only one sample)
repeated measures t test assumptions
- each score is independent
- difference scores are normally distributed
- NOT homogeneity
t or f: repeated measures t test are more economical but have a lower power
false, they are more economical and they have higher power
pros of repeated measures t tests
- more economical
- higher power
- no carry over effects
- less vulnerable to demand characteristics
demand characteristics
- things in experiments that lead participants onto what the researcher is attempting to study
- can make good or evil subjects if they determine the hypothesis (both bad in the end)
what test is used to compare two means from the same population
paired sample/repeated measures t test
scale vs ordnial vs nominal numbers
- scale: theoretically infinite amount of numbers, equal intervals, cont., #s are meaningful
- ordinal: categorical, no positions/order, meaningful but not measurable difference (ex always/sometimes/never
- nominal: discrete/categorical, not smth you can measure difference of (ex course codes)
