Psyc 305 Stats

Front 1

nominal vs. ratio variable

Back 1

has a name to it (species) vs. has a number to it (weight)

Front 2

type I vs. type II error

Back 2

alpha (false alarm) vs. beta (miss) error

Front 3

forumla for variance and deviance

Back 3

variance = (x-mean)^2/N-1

deviance = sqrt(variance)

Front 4

parameters

Back 4

properties of populations

Front 5

in the ANOVA, are variables nominal or ratio?

Back 5

- the dv is ratio

- the ivs are nominal

Front 6

sampling error

Back 6

diff between an estimate and the corresponding parameter

Front 7

univariate vs. multivariate

Back 7

- 1 dv vs. multiple dvs.

Front 8

counterbalancing

Back 8

- presenting a random order of the levels of a repeated measures factor to each subject experiencing the repeated measure

Front 9

assumption made by using MS (what, how)

Back 9

- assumption of homogeneity of variance (especially for pooled variance)
- cause it averages the variance of subject scores across groups

Front 10

expected mean squares (definition, what they tell us)

Back 10

these try to predict what the mean square for a certain source looks like in the population. you derive the source of variation from these.

they tell us that F has to be at least >1 for it to be a significant result.

Front 11

the assumptions underlying ANOVA (4)

Back 11

- that subjects are randomly sampled (sample represents the population)

- independent observations (that each observation has no influence on the next)

- normal distribution (gaussians for each factor)

- homogeneity of variance (group variances are equal (no skew) )

Front 12

the Fmax test (what for, formula)

Back 12

- testing the assumption of homogeneity of variance. usually when there are outlying data points or unequal n's.

- Fmax = s^2 largest group variance / s^2 smallest group variance

If ABOVE critical value, then it IS VIOLATED and TRANSFORMATIONS NEEDED

Front 13

error term rules (3)

Back 13

1. if a is crossed with no random factors, the error term is at the bottom of the anova table

2. if a is crossed with 1 random factor s, the error term is MSas

3. if a is crossed with more than 1 random factor, there is no simple error term

Front 14

why are unequal n's bad?

Back 14

- because they make A and B related and not orthogonal anymore

- when one of the n's is 3 times bigger than the other, then it throws off the calculation so bad that you cant even perform the anova

Front 15

homogeneity of covariance

Back 15

an assumption that a repeated measures design makes.

it basically means that the levels of the repeated measure are all equally variant from from each other. this gets harder to prove the more and more levels you have, which is why you apply the Greenhouse Geisser conservative df correction for significant results. for results that are at first significant and then become insignificant, you can apply another correction through the computer.

GG is a very conservative test. it prevents alpha error (type 1)

Front 16

advantages and disadvantages for repeated measures designs (4)

Back 16

advantages
-require fewer subjects
-smaller error terms - large the variation between subjects, the smaller the left over error terms

disadvantages
- carry-over effects (can reduce with counterbalancing)
- can violate the assumption of covariance

Front 17

scheffe test critical value

Back 17

Fc = k-1( Fc(k-1, df))

Front 18

why use tests of simple main effects?

Back 18

- because each post hoc gets weaker (except for t-tests) as the # of means being compared increases. this is cause the other posthocs all use setwise correction, and the critical values of the test can be so high that you dont find significance.

- simple main effects allows you to isolate the row/column that is the source of variation before entering post-hocs, reducing the # of means you have to compare for post hocs.

Front 19

when are the two cases to use planned comparisons?

Back 19

- when you want to know something specific about the ANOVA (make a specific comparison) and dont care about the majority of the results

- to test the difference between two non-significant results

Front 20

what happens if there is non-orthogonality in the planned comparison?

Back 20

- you basically carry through with the calculation for SS, MS, and F.

-when choosing a Fcrit, do a Bonferroni correction. Make sure the alpha level is divided by the # of comparisons you are making

Front 21

assumptions of a planned comparison

Back 21

that the factor you are testing is nominal and equally spaced apart