Research Module 7: Correlations

Question 1

what tests can you run when you’re looking for differences?

Answer 1

t-test
ANOVA
MANOVA

Question 2

what tests can you run when looking for relationships?

Answer 2

correlation (bivariant)

ICC (intraclass correlation coefficients)

regression

internal consistency

Question 3

inter-rater reliability

Answer 3

relationship between different raters scores

Question 4

intra-rater reliability

Answer 4

relationship between the 2 sets of scores from the same rater

Question 5

test-retest reliability

Answer 5

The degree to which a test yields similar results when it is administered to the same person (or group) on two or more separate occasions.

Question 6

internal consistency

Answer 6

are the items in a test or survey consistent with each other

Question 7

predictive validity

Answer 7

can a score on one outcome measure predict another outcome measure

Question 8

concurrent validity

Answer 8

can 2 different outcome measures get similar results

Question 9

criterion validity

Answer 9

can 2 different outcome measures (one is a gold standard) get similar results

Question 10

Pearson correlation

Answer 10

“is there a relationship between 2 variables and how strong is it?”

commonly used in methodological research

must have continuous data

Question 11

construct validity

Answer 11

can subgroups of items in an outcome measure explain factors of the same complex construct (such as satisfaction, coordination, etc)

Question 12

r

Answer 12

statistical abbreviation for a Pearson correlation

shows the probability of the relationship emerging by chance

represents the strength of the relationship

Question 13

r = +1

vs

r = -1

Answer 13

perfect positive correlation

perfect negative correlation

Question 14

if the p-value is 0.0032, how do you interpret this?

Answer 14

there is only a 0.32% chance that the relationship occurred just due to chance

this is a “real” relationship

<0.05 is significant

Question 15

Answer 15

positive association

increased motivation, increased GPA

Question 16

Answer 16

negative association

as students number of absences decrease, GPA increases

Question 17

what is similar and different between these 3 graphs

Answer 17

similar: same r

different: slopes

Question 18

what is the relationship between r and slope?

Answer 18

THEY ARE NOT THE SAME THING

slope: tells you how changing 1 unit is related to another variable

Question 19

Answer 19

r = 0.7

strong positive correlation

Question 20

Answer 20

r = 0.3

weak positive correlation

Question 21

Answer 21

r = 0

no correlation

Question 22

Answer 22

r = -0.7

strong negative correlation

Question 23

Answer 23

r = -0.3

weak negative correlation

Question 24

Answer 24

r = 0

no correlation

Question 25

Answer 25

r = 1 strong positive correlation

Question 26

Answer 26

r = -1 strong negative correlation

Question 27

r = 0 - 0.19

Answer 27

very weak positive correlation

Question 28

r = 0.20-0.39

Answer 28

weak positive correlation

Question 29

r = 0.4-0.59

Answer 29

moderate positive correlation

Question 30

r = 0.6-0.79

Answer 30

strong positive correlation

Question 31

r = 0.8-1

Answer 31

very strong positive correlation

Question 32

coefficient of determination

Answer 32

how much of the variability in one variable can be predicted by the other variable r^2

Question 33

what does a correlation of 0.845 mean for r^2?

Answer 33

r^2 = 0.714 meaning 71.4% of the variability in device B can be predicted by device A (and vice versa)

Question 34

is the following scenario a useful relationship: r = 0.205

Answer 34

no - while it is statistically significant, only 4% of the variance is shared because r^2 = 0.04 = 4%

Question 35

what is the interpretation of r = 0.8

Answer 35

r^2 = 0.64 = 64% of variable Y can be predicted by variable X and vice versa

Question 36

Pearson's correlation measures the __________ between 2 variables, NOT the _____________.

Answer 36

relationship agreement

Question 37

agreement

Answer 37

the degree to which different methods, observers, or instruments produce the **same results** when measuring the same variable.

Question 38

what does this mean: there will be a significant relationship between time 1 and time 2 at r>+0.8 or r<-0.8

Answer 38

ROM recorded at test 1 can predict 64% of the variance at time 2

Question 39

interpret the following: r = 0.952 p<0.001 95% CI: 0.894 - 0.979

Answer 39

strong significant positive relationship 90.6% of time 2 scores can be predicted from time 1 scores 95% of the time, the true correlation will be between 0.894 - 0.979

Question 40

point-biserial correlation (r^pb)

Answer 40

special case of Pearson correlation that is run between 2 levels of a **categorical variable** (ex: 2 groups) and an **interval/ratio variable**

Question 41

what correlation test would you run for the following: is there an association between bachelor's trained nurses and master's trained nurses and their average patient satisfaction scores at Methodist hospital?

Answer 41

point biseral correlation in this case, the patient satisfaction score is the continuous variable and the bachelors/masters trained nurses are the categorical variable (2 levels)

Question 42

spearman coefficient

Answer 42

nonparametric equivalent to the Pearson's correlation used with **ordinal** data

Question 43

what correlation should you run for the following: you are interested in childhood aggression (the relationship in mobility scores and social skills scores in formally premature children). is there a relationship between these 2 sets of skills or is it random?

Answer 43

spearman (rs) the mobility and social skills are scored using ordinal data

Question 44

for a spearman correlation, the larger the difference, the _________ the association

Answer 44

smaller

Question 45

if the rs is 0..769 and p = 0.044, what does this mean

Answer 45

there is a strong, positive significant relationship, and the chance that this was just random chance is 4.4%

Question 46

internal consistency

Answer 46

how closely related the items in an outcome measure are as a group surveys: how good are the questions

Question 47

Cronbach's α

Answer 47

**internal consistency statistic** that is calculated from averaging all of the possible pairwise correlations between items and takes into account the number of items and variance too It tells you how well a group of items measures the same underlying concept.

Question 48

Cronbach's α interpretation

Answer 48

0.7: acceptable internal consistency 0.8: good 0.9: excellent

Question 49

standard error of the mean (SEM)

Answer 49

value that describes the difference between the sample mean and the true population how far the sample mean of data is likely to be from the true population mean

Question 50

standard deviation

Answer 50

measures the amount of variability, or dispersion, for a set of data from the mean

Question 51

relationship between SEM and SD

Answer 51

SEM < SD

Question 52

standard error of the mean formula

Answer 52

measures the average distance between the M (sample M) and u (population mean) how accurately a sample mean represents its corresponding population mean SEM = σ/√ n

Question 53

standard error of measurement (SEm)

Answer 53

related to test reliability gives you an idea of how much error you should expect from a measurement

Question 54

standard error of measurement formula

Answer 54

SD√(1-r) r=test-retest reliability

Question 55

what is the difference between standard deviation and standard error of measurement

Answer 55

SD: **variability** within a sample SEm: how **precisely** you've estimated the mean score

Question 56

how would you interpret a SD = 5, r=0.8, and SEm = 2 if a patient scores a 10

Answer 56

based on the SEm of 2, 68% (1 SEm = 2) of the time this patients true score is between 8 and 12 96% of the time, the person's true measurement will fall within 2x the SEm (or 6 to 14). SEm = 5√(1-0.8) = 2 1 SEm = 2 so 10 +/- 2 = 8 and 12 2 SEm = 2x2 = 4 so 10 +/- 4 = 6 and 14

Question 57

intraclass correlation coefficient (ICC)

Answer 57

commonly used to calculate **inter-rater reliability** of more than 2 raters, like a Pearson but with more flexibility Pearson: only 2 sets of scores ICC: **many sets of scores**

Question 58

how to interrupt an ICC score

Answer 58

0-1...1 being then most consistent

Question 59

how are ICC's written in articles?

Answer 59

ICC(model, form)

Question 60

ICC model 1

Answer 60

each participant is assessed by a different set of randomly selected raters (this is rare) Each subject is measured by a different, random rater, and there’s no overlap. Example: Each patient is rated by a different physical therapy student.

Question 61

ICC model 2

Answer 61

each participant is assessed by each rater, and raters have been randomly selected and represent all similar raters (this is common) All subjects are rated by the same raters, and those raters are randomly selected representatives from a larger population. ex: 3 randomly selected physical therapists all measure each patient’s ROM.

Question 62

ICC model 3

Answer 62

each participant is assessed by each rater, but the raters are the only raters of interest (this is somewhere in the middle) All subjects are rated by the **same specific raters**, and you only care about those exact raters. Example: You have 20 patients assessed by two lead therapists in your clinic using a specific post-op knee protocol.

Question 63

ICC form 1

Answer 63

single measurement (1 rating from each rater)

Question 64

ICC form 2

Answer 64

average of 2 measurements (two ratings from each rater)

Question 65

ICC form 3

Answer 65

average of 3 measurements (3 ratings from each rater)

Question 66

what does SPSS call ICC models

Answer 66

model 1: one way random model 2: two way random model 3: two way mixed

Question 67

what correlation would you run for the following scenario: using 4 athletes completing one standing long jump, 3 raters, each rater is asked to use the stopwatch on their smart phone to measure jump time

Answer 67

more than 2 raters → ICC (2, 1) model 2: raters choose at random, representative of all similar raters form 1: assessments per rater per person = 1