Module 10, Hypothesis Testing Two Samples

Question 1

Between Group Comparisons:

Answer 1

enable us to determine if a difference between two groups is large enough to be statistically significant, in order to reject the null

Question 2

Within Groups Comparisons:

Answer 2

deals with the effects of experimental interventions, comparing groups before and after the intervention, LIKE A WITHIN SUBJECTS MEASURE

Question 3

Standard Error of the Mean

Answer 3

Measure of variability of sample means around the population mean in a sampling distribution of the mean
SD of the sampling distribution

Question 4

Degrees of Freedom pooled

Answer 4

When two independent samples with approximately equal variances the df are determined by summing the sample sizes and then subtracting the number of samples
Df pooled = N1 + n2—2

Question 5

Order of subtraction

Answer 5

Order of Subtraction
Alternative hypothesis determines the order of subtraction
Subtract the one that is not specified to be higher from the one that is

Question 6

Independent Samples/ Between Subjects Design

Answer 6

Randomized, Independent groups, between subjects ALL MEAN THE SAME THING
Don’t always have to compare an experimental group to a control group: common misconception is that experiment has to have control condition THIS ISN’T TRUE; sometimes not possible to have a control group
Can compare two different amounts of an IV (compare one dose to another dose of drug when you already know it works)
Can be an experiment or ex post facto: two groups that are preexisting also works
Each participant participates in ONE AND ONLY ONE GROUP
Comparisons made between group
RANDOMIZED DESIGN: Random assignment used to assign participants to groups (IF IT IS A TRUE EXPERIMENT)

Question 7

Levene’s Test for Equality of Variance

Answer 7

if significant, then variances are not equal

Question 8

Conditions of Independent T Test

Answer 8

Each population should have a normal distribution AS LONG AS SAMPLE SIZE IS OVER 25
The samples must be independent: one participant’s score does not influence another score
Homogenity of variance: both groups must be sampled from populations with similar variance

Question 9

Effect size interpretations, uses of effect see

Answer 9

D= 0.2 (SMALL)
d=0.5 (Medium)
d= (0.8) LARGE
Compare how meaningful an effect is between studies without knowing the exact scale they used: ratio of difference to variability
Allows us to determine our ability to detect a difference,s hould one exist for a particular sample size, if we have an estimate of the effect size
We use our best guess of the effect size with the sample size we have ^^

Question 10

Delta:

Answer 10

value used in referring to power tables that combine effect size and sample size
Power is always between 0 and 1, can never be 1 because you can never be 100% sure that you’re going to find an effect

Question 11

How does increasing affect size increase power?

Answer 11

Increase effect size: difference between null and alternative is effect size
Move distributions
Make this manipulation stronger: but might compromise external validity/may not be practical/ethical

Question 12

Advantages and disadvantages of increasing sample size on power

Answer 12

WORKS BECAUSE OF THE CLT: standard error goes down as n increases
Greater sample size = skinnier distribution (less variability/spread)
Disadvantages: may be difficult to recruit a large sample, much more expensive with time, money, lab equipment
STILL THE BEST APPROACH TO INCREASING POWER; DOESN’T JEOPARDIZE EXTERNAL OR INTERNAL VALIDITY (doesn’t increase chance of making Type I error)

Question 13

Diusadvantages and advantages of decreasing variance on power

Answer 13

Decrease population variance: decreasing “noise of participants” increases power
Decrease variance: more sure about our estimate of the true parameter if everything is less variable
Make distributions skinnier, decreases variance, and increases power
Getting distributions to be skinnier: use less diverse population=homogenity, helps to detect effect
DISADVANTAGES: results may not generalize properly, because you do not have a diverse sample
May be hard to find people with that particular characteristic you’re looking for from the population

Question 14

Effects of using a one tailed vs two tailed test on power

Answer 14

Use a one-tailed test instead of two tailed: move the criterion line on the one side, same chance of making TYpe I error but now putting it all in one tail: DECREASES BETA AND INCREASES POWER (only use this if youre very sure of the direction of your effect), if you’re not sure you shouldn’t use it, and it goes against convention (leads people to believe you originally did a two tailed test and it didn’t work so then you changed it to do a one tailed test, could manipulate the data dishonestly**) can fix this with pre-registering your hypothesis before you collect data

Question 15

Effects of changing alpha on power

Answer 15

Could change the alpha level, decreases the threshold needed for significance: BUT INCREASES CHANGE OF MAKING TYPE ONE ERROR (unfortunately) but also increases ability to find effect (usually don’t want to do this)

Question 16

Matched Pairs

Answer 16

When there is a lot of variance in a sample, using MATCHED PAIRS is effective to combat it: then randomly assign people in that pair to one condition or another
Pairing on similar variables that you think might influence the outcome then randomly assigning within those pairs is useful
Could be experimental or quasi experimental (similar in age/gender etc.)

Question 17

Natural pairs Quasi design

Answer 17

Natural Pairs: when participants are matched based on some biological or social relationship that IS PREXISTING: could recruit pairs of siblings (for example studying personality characteristics influencing alcoholism when controlling for other things such as genetic variables)–solves the issue of isolating personality characteristics while controlling for genetic and environmental factors
Ideally use it when equating participants on a number of potential extraneous factors
These types of studies are typically quasi experimental
Getting enough of these pairs, because they’re equated on other variables, might help you get a better result

Question 18

Benefits and cons with repeated measures

Answer 18

Repeated Measures: don’t need as many participants, every individual experiencing all levels of the IV, reduces variance
IDEAL in matching
Carryover effects are a problem (like with drug dosages), sometimes can’t measure an outcome multiple times (like taking a test multiple times, eliminates the point because of practice/testing effects)
Ethical Consideration: whether participants can cope with repeated testing , if its invasive or stringent
Comparing measurement of one person to a second measurement of themselves, or a measure to their matched participant (become comparisons for each other)

Question 19

Dependent paired vs independent

Answer 19

Dependent Paired: greater control over the equality of the group, have ensured equality on some variables beforehand, more statistical power to find an effect because variability due to individual differences is NONEXISTENT in within subjects and decreased in matched pairs design, allows for smaller sample size
Independent: much easier to conduct, random assignment is the only thing that’s needed, need an increased n here, not always possible to conduct within-subjects, if you have enough people the statistical advantage that within-groups has becomes less meaningful

Question 20

Paired T test assumptions

Answer 20

Assumptions: the samples must be dependent (paired with another score), both populations must be normally distributed
If these conditons are met, we’re using the null hypothesis and degrees of freedom n – 1
Degrees of freedom= n–1

Question 21

what does the numerator of the f statistic represent?

Answer 21

The numerator of the F statistic represents the effect of the Iv

Question 22

what do z t and f calculations have in common?

Answer 22

numerator of the test statistic contains a measure of difference between means

Question 23

Can f ever be negative

Answer 23

no; this would indicate calculation error