Stats Uni Year 2

Front 1

Is the fit of a model better or worse when the squared error is smaller

Back 1

Better

Front 2

3 ways to enter predictors into a model

Back 2

Hierarchical

Forced entry

Stepwise

Front 3

What is stepwise

Back 3

When predictors are entered into a model based on a criteria (semi-partial correlation with the outcome)

Front 4

Issue with stepwise

Back 4

Can produce spurious results (false/fake)

Can be arbitrary which predictor should go in first, bad decision at the start can spiral

Getting dressed example

Front 5

What does standardizing parameter estimates allow for?

Back 5

Comparison between them, see which has more of an effect (have to look at the b values and actual context tho)

Front 6

What's another name for mean squared error?

Back 6

Variance

Front 7

What's the equation for the mean squared error?

Back 7

Front 8

What is SST

Back 8

Total variability (between scores and the mean - whole cake)

Front 9

What is SSR

Back 9

Residual/error variability (half of cake)

Front 10

What is SSM

Back 10

Model variability (dif in variability between the model and the mean, part of cake)

Front 11

What is the F ratio

Back 11

Ratio between SSR and SSM

MSM / MSR

Front 12

What is R^2

Back 12

The proportion of variance accounted for by the model

Correlation between observed and predicted scores

How big is the slice of cake relative to the model in relation to the whole cake, what proportion of the original variability is the model explaining by fitting it

SSM/SST

Front 13

What is the adjusted R^2

Back 13

An estimate of the R^2 in the populations

Front 14

What is adjusted R^2

Back 14

An estimate of R^2 in the populations

Front 15

What are confidence infervals

Back 15

Intervals that contains the 'true' population value of the parameter in 95% of samples

Front 16

How big does n need to be to assume the sampling distribution is normal?

Back 16

About 150

Front 17

How to calculate the confidence interval

Back 17

Lower bound = Mean - 1.96 × SE

Upper bound = Mean + 1.96 × SE

Front 18

What is a type 1 error

Back 18

Reject a true null hypothesis

Not pregnant but say they are

If 0.05 gets bigger, increases error

False positive

Hint 18

Positive or negative comes first?

Front 19

What is a type 2 error

Back 19

Failure to reject a null hypothesis that is actually false. Reject alternative hypothesis but it does not occur due to chance

If 0.05 gets smaller, error increases

False negative

Hint 19

What comes first, negativity or positivity?

Front 20

4 ways to see the effect size

Back 20

The parameters, b

Standardized b

Pearsons r

Cohen's d

Front 21

Pearsons r effect size cut offs

Back 21

0.1 = small

0.5 = large

Front 22

Cohen's d effect size cut offs

Back 22

0.2 small

0.8 large

Front 23

How to calculate Cohen's d

Back 23

Difference in means divided by SD (pooled or control)

Front 24

How to calculate the SD pooled

Back 24

Front 25

What are these called?

Back 25

Influential cases

Front 26

3 ways to detect outliers and influential cases

Back 26

Graphs

Standardized residuals

Cooks distance

Front 27

Above what value is a standardized residual likely to be an outlier?

Back 27

3 or above

Front 28

Above what value of cooks distance is cause for concern?

Back 28

1 and above

Front 29

What are the key assumptions of the linear model

Back 29

Linearity and additivity

Spherical residuals (Homoscedasticity errors, independent errors)

Normality (residuals)

Sampling distribution

Front 30

What is linearity and additivity

Back 30

Relationship between predictor and outcome is linear

The combined effect of predictors is additive

Front 31

How to test linearity and additivity

Back 31

Look at graphs

Front 32

What does independent errors mean

Back 32

The error is prediction (residulas) for one case should not be related to the error in prediction for another case (autocorrelation)

Front 33

What does homoscedasticity mean

Back 33

Varience of errors (residuals) consistent at different values of the predictor variable

Front 34

What does it mean if the spherical errors assumption is violated?

Back 34

b values are unbiased but not optimal

Standard errors are incorrect meaning t-tests, p values and confidence intervals will also be incorrect

Front 35

How to detect spherical errors

Back 35

Graph of standardized residuals: ZRESID against standardized predicted ZPRED

Front 36

What is heterogeneous variance

Back 36

Where variances are different at different points

Front 37

In the GLM, does the data have to be normally distributed

Back 37

No, normality if residuals and Sampling distribution important

Front 38

In the GLM, does the data have to be normally distributed

Back 38

No, normality if residuals and Sampling distribution important

Front 39

If the residuals are not normally distributed, what does this mean?

Back 39

Doesn't really matter

b will be unbiased and optimal (I.e. will minimize the variance) but there may be classes of estimator that are more accurate

Front 40

Why is normality of sampling distribution important?

Back 40

P values associated with bs of the model assume the test statistic associated with them follows a normal distribution

Also, confidence intervals for bs are constructed using a quartile from a null distribution assumed to be normal

Front 41

If residuals are not normally distributed, what do you use

Back 41

Central limit theorem, increased sample size makes it more normal, issues of normality goes away

Front 42

What graphs can be used to explore normality?

Back 42

Histograms

Boxplots

P-P/Q-Q Plots

Front 43

Above what value shows skew

Back 43

2.5

Front 44

What is the Kolmogorov-Smirnov (K-S) test

Back 44

A bad test of normality

Front 45

4 robust estimations for correcting problems

Back 45

20% trim

M-estimators

The bootstrap

Adjust SE or test statistic for heteroscedacity

Front 46

Method of the bootstrap

Back 46

Gets a sample from your sample

Gets value, puts it back, etc until the same n as your sample

Repeats 1000 times

Order value and work out which fall in 95%

Gives a CI based in the data of the sample not population, no issues in normality

Front 47

What does the F statistic show

Back 47

If the predictor help to improve our ability to predict the outcome

Front 48

Equation for F

Back 48

F = MSM/MSR

Front 49

In dummy coding, what does b0 mean?

Back 49

The mean of the control or 'zero coded' group

Front 50

In dummy coding, what does b1 mean?

Back 50

The difference between the 2 variables

Front 51

How do you test the difference between two groups to see if its significant?

Back 51

T-tests

Front 52

Equation for the mean squared error

Back 52

MS = SS/df

Front 53

What 2 things can correct Heteroscedacity

Back 53

Welch

Brown-Forsythe

Front 54

2 reasons to do contrast coding over dummy coding

Back 54

Control the error rate - make sure its 5% across all sig tests across all of model

Want to override what SPSS automatically gives you

Front 55

What is another name for contrast coding

Back 55

Orthogonal contrast

Front 56

Opposite test to contrast coding

Back 56

Post hoc tests

Front 57

Is the contrast coding planned/hypothesis driven

Back 57

Yes

Front 58

What are post hoc tests

Back 58

Multiple t-tests with adjusted p-values

Front 59

When you have ordered means, what could you use to compare them?

Back 59

Trend analysis

Front 60

Why is dummy coding error rates not independent

Back 60

Uses the control, or category 0 more than once

Front 61

How many contrast should there be?

Back 61

K-1

Should always end up with 1 less contrast than the number of groups

Front 62

What are the 4 rules to coding planned contrasts

Back 62

1. Groups coded with a +ve vs groups coded with a -ve

2. Sum of weights for a comparison should be 0

3. If a group is not involved in a comparison, assign it a weight of 0

4. For a given contrast, the weights assigned to the group(s) in 1 chunk of variation should be equal to the number of groups in the opposite chunk of variation

Front 63

General linear equation for the contrast model

Back 63

Y = bhat 0 + bhat 1(contrast 1) + bhat 2(contrast 2) + E

Front 64

Are constrast coding or post hoc tests better n why?

Back 64

Contrast coding, much better to be theory driven and design and test theories in a scientific way

Front 65

What is the problem with post hoc tests

Back 65

Inflates the type 1 error rate, comparisons of all means

Increases chances of false positive

Front 66

Solution to the issue of post hoc issue

Back 66

Adjust the alpha (or test statistic) to be more conservative

Bonferroni

However/ lose power to detect differences

Front 67

Name for comparing means adjustong for other predictors using a linear model

Back 67

Analysis of covariance ANCOVA

Front 68

2 advantages of ANCOVA

Back 68

Can reduce error variance (explains some SSR)

Greater experimental control (gain greater insight into the effect of the predictor variable)

Front 69

When do you bootstrap

Back 69

When there's no normal distribution/sample is small

Front 70

What is a covariate?

Back 70

A continuous predictor

Front 71

What does it mean if a covariate shows heterogeneity of slopes

Back 71

Relarionship between predictor and covariate is not consistent across groups - cannot interpret the effect of the covariate as it changes between predictors

Front 72

What is SPSS's name for dummy coding

Back 72

Simple contrast

Front 73

What means should you look at if including a covariate

Back 73

Adjusted

Front 74

If the interaction is significant between IV and covariate, what does that mean

Back 74

Assumption of homogeneity broken, BAD

Front 75

What is a factorial design

Back 75

More than 1 IV/predictor variable that has been manipulated

Front 76

Benefit of factorial designs

Back 76

Can look at moderation interactions effects

Front 77

Equation for factorial design

Back 77

Outcome = b0 + b1 predictor + b2 moderator + b3 interaction + E

Front 78

If there is a significant interaction, what do you consider about the main effect?

Back 78

Ignore it

Front 79

Benefits of repeated measures design

Back 79

Sensitivity (unsystematic variance is reduced, more sensitive to experimental effects)

Economy (less ppts needed)

Front 80

What assumption does a repeated measures design violate?

Back 80

Independent residuals

Front 81

How do you correct for repeated measures breaking the assumption of independent residuals

Back 81

Impose extra assump - the assumption of sphericity

Estimate degree it breaks, then adjust the DoF by amount it violates the assumotion

Front 82

What is the assumption of sphericity

Back 82

Means the correlation between treatment level is the same

Assumes variances of differences between conditions is equal

Front 83

The assumption of sphericity is adjusted for using what 2 things

Back 83

Greenhouse-Geisser estimate

Huynh-Feldt estimate

Front 84

If the GG or HF estimates are 1, what does this mean

Back 84

Data is perfectly spherical

Front 85

What do you multiply the GG or HF estimates by to correct for the effect of sphericity?

Back 85

df

Front 86

What is the difference between GH and HF

Back 86

GG conservative (safer)

HF liberal

Front 87

What does the lower bound estimate show in assumption of sphericity

Back 87

Lowest value possible value that sphericity could be, given the data

Front 88

What are the follow-up tests you can do after repeated measures

Back 88

Built in contrasts (e.g. simple contrasts)

Post hoc tests: H/ only 3/18, bonferroni

Front 89

What does the bonferroni test do

Back 89

Correcting test statistic so across all tests you do, the alpha rate never rises above 5% as thats what you have it set as

Front 90

What does ANOVA stand for?

Back 90

Analysis of Variance

Front 91

If you have categorical outcomes, what do you predict?

Back 91

Log odd of the outcome occuring

Front 92

What is b when you have categorical outcomes

Back 92

The change in the log odds of the outcome associated with a unit change in the predictor

Front 93

If b1 is less than one in categorical outcomes, what does this mean?

Back 93

Predictor ^, prob of outcome occuring v

Front 94

If b1 is more than one in categorical outcomes, what does this mean?

Back 94

Predictor ^, prob of outcome occuring ^

Front 95

Methods to add parameters to the model in categorical outcomes

Back 95

Forced entry (variables added simultaneously)

Hierarchical (enterd in blocks based on past research)

Stepwise (SPSS based on stat criteria)

Front 96

2 unique problems with categorical outcomes

Back 96

Incomplete info (inflates SE)

complete separation (when the outcome variable can be perfectly predicted, messes up SE, model completely nonsensical)