Four Steps of Hypothesis Testing
state hypothesis
critical values
Collect and calculate data
make decisions
what must one always remember when writing their hypothesis statement
always write it in terms of the population
anti-hypothesis; no treatment effect or significant differences took place
null hypothesis (Ho)
treatment will cause a change or significant difference; there can be multiple
alternative hypothesis (H1)
looking for values that fall into the extreme 5%,1, or .1% of scores
alpha level
composed of the extreme values that are very unlikely to be obtained if the null hypothesis is true
critical region
what happens when critical region becomes smaller
values are more towards extreme ends of distribution
what happens as alpha level decreases
probability of type I error happening decreases
two tailed vs one tailed
two tailed is a nondirectional test because it doesn't tell us what direction the difference is
one tailed is directional test because it tells us whether there is an increase or decrease
Standard Error Formula
Om = o/√n
Z Score formula for sample mean
Z= m-M/Om
what do we do when we answer yes to step four of the hypothesis method
reject the null
what do we do when we answer no to step four of the hypothesis method
fail to reject null
as sample size increases
standard error (Om) decreases
Cohen's D
mean difference/ standard deviation
small Cohen's D
d ≤ .2
medium Cohen's D
.2 ≤ d .≤ 8
large Cohen's D
d ≥ .8
explain the term most conservative
-score that is furthest from the mean -by choosing z score that farther from mean, you're reducing likelyhood of committing type I error
when you have a one tailed directional test you're more likely to
reject the null
when given the opportunity, one should more likely pick a
one tailed test bc you're less likely to commit a type II error
Type I Error
-determined by ALPHA LVL -considered worst type of error -incorrectly rejected a true null -when one claims there's a treatment effect but in reality there is not -ex. sending an innocent person to jail
Type II Error
-failed to reject a true null -we thought there was not treatment effect but there was -determined by calculating BETA, works w/ Power -ex. setting a guilty person free
tells one the likelihood that one will correctly reject a false null
Power
how do we determine Beta
1-Power= Beta
as mean difference increases
test statistic and rejecting the null both increase
as mean difference decreases
test statistic and rejecting the null both decrease
as standard deviation(o) and standard error (Om) both increase
test statistic and rejecting the null both decrease
as standard deviation(o) and standard error (Om) both decrease
test statistic and rejecting the null both increase
as sample sizes increases
standard error (Om), test statistic, and rejecting the null all increase
Assumptions of Z Score Testing
Random Sampling: independent random sampling & random sampling with replacement
Independence of Observations: each individual's score is independent of each other, it doesn't impact the other
Value of Pop. Standard Dev. isn't changed by treatment: how we calculate error
Normal distributions: is sample size is 30 or greater or if it comes from normal distribution population
reporting results for a z statistic
"In a sample, n=6, it was found that (fancy words), z= +3.40, p<.05".
Single Sample T test
when we don't have population variance we use sample variance
estimated standard error formula
Sm= √ s2/n
Formula for Single Sample T test
t= m-M/Sm
When do we use a Z score and when do we use a T test
-if you have population standard dev -> Z score -if you don't have population standard dev. & working with single sample -> T test
when sample statistic doesn't accurately affect population parameter
biased statistic
Larger the sample size and degrees of freedom
more closely proportionationated will those scores math to the Unit Normal Table
smaller the sample size and degrees of freedom
more flattened out the t distribution will be
sample variance formula
s2= ss/df
percentage of variance explained
-r2 -percentage of variance due to treatment
r2 formula
r2= t2/ t2 + df
small r2
0.01 < r2 < 0.09
medium r2
0.09 < r2 < 0.25
large r2
r2 > 0.25
results statement for a single sample t test
" The participants averaged m=5 with a SD=4 on the happiness scale. Statistical analysis found that (more fancy words I should be asleep right now rrrriiiiiipppppppp), t(8-df)=3.00, p<0.5, r2= 24.56%'.
Assumptions about single sample t test
independence of observations: scores are independent thus they don't affect each other
normal population distributions: sample size of 30+ or comes from normal population distribution
Independent measures of between subjects t test
-two separate samples and we're comparing the means together -big clue if you have two different data sets ( n1,n2,SS1,SS2)
stating the hypothesis for a Independent measures of between subjects t test
null: M1-M2 = 0 alt: M1-M2 ≠ 0
DF for Independent measures of between subjects t test
since you have two samples, you add the two DF DF= df1 + df2
Pooled Variance Formula
S2p= SS1 + SS2 / df1 + df2
estimated standard error formula that we learned for CH10 when we have two samples and two sample means
Sm1-m2= √ S2p/n1 + S2p/n2
Formula for Independent measures of between subjects t test
t= m1-m2/ Sm1-m2
effect size => estimated Cohen's D
when the mean difference is negative, you take the absolute value ex. -7.5 becomes I 7.5 I
when we state with a certain degree of assurance that the actual population mean difference fall within a certain range
confidence interval
confidence interval percentages can be
50%, 80% or 90%
confidence interval formula
M1-M2= m1-m2 ± (tCI)(Sm1-m2)
result statement for Independent measures of between subjects t test
"Have you drank water today, didn't think so go do it, t(3)=2.50, p<.05, d=56.7, r2= 56.64%, CI=[ 2.14, 9.18]".
Assumptions of Independent measures or between subjects t test
independence of observations: scores independent of each other, they don't affect each other
normality: sample size of 30+
Homogeneity of variance: variance of population from which samples occur are the same ;Levine statistic; Hartley's Fmax
test to asses and make sure population from which samples come from have the same variance
Levine statistic
requires us to have equal sample sizes
Hartley's Fmax
Repeated Measures T Test
-we have one sample but experiment the same sample under two separate conditions -remove all individual differences thus makes it easier to pinpoint the actual cause of change -comparing different scores -ex. taking a test one day in a 60degrees room vs another day we take it in a 90degrees
Matched Subjects design
-we have two samples but try to make them as identical as possible -ex. matching up the kids with similar characteristics taking a test and grouping them together to make it look we have one sample but we really have two
difference of scores formula
X2-X1
mD formula
sum of difference of scores/ sample size
writing a hypothesis for a repeated measures or matched subjects design
null: mD= 0 alt: mD ≠ 0
estimated standard error with a D subscript formula
SmD= √ s2/n
Formula for repeated measures or matched subject design t test
t = mD/SmD
we use the ≤ or ≥ symbol when
we have one tailed
we use the = or ≠ when
we have two tailed
confidence interval formula for MD
MD= mD ± (tCI)(SmD)