Categorical Data Analysis

Question 1

Chi-squared assumptions

Answer 1

• All expected values Ei,j are greater than the value of 1.
  
  • No more than 20% of the expected values Ei,j are less than 5.
  • Cell values are independent.

Question 2

test for a 2x2 table

Answer 2

Chi-squared or Fisher’s

Question 3

Test for 2x3+ table

Answer 3

Chi-squared test for trend

Question 4

Test for paired data

Answer 4

McNemar’s

Question 5

Test for a lurking/stratifying variable

Answer 5

Cochran Mantel Haenszel Test

Question 6

When do you use a Fisher’s exact test?

Answer 6

when the assumptions of a Chi-squared are violated

Question 7

How must data be formatted for Fisher’s exact?

Answer 7

must be a 2x2 table
data may need to be combined

Question 8

Fisher’s hypotheses

Answer 8

H0: There is no association between the variables (independent)
H1: There is an association between the variables (dependent)

Question 9

What is risk (and how to calculate)?

Answer 9

> the probability that an event will occur
number of events/total population at risk

Question 10

risk ratio or relative risk calculation

Answer 10

p(event in group 1)/ p(event in group 2)

Question 11

odds definition and calculation

Answer 11

odds is the ratio of an event happening to not happening
odds = p / (1-p)
where p is the probability of an event

Question 12

what is an odds ratio?

Answer 12

a measure of association between an exposure and an outcome

Question 13

odds ratio calc

Answer 13

odds of event in exposed group/odds of event in non-exposed group

Question 14

odds ratio interpretation

Answer 14

the exposed have x times the odds of the event occurring than the non-exposed

Question 15

odds ratio table set up

Answer 15

event along the top (columns)
exposure along side (rows)
yes + yes in top left

Question 16

when is relative risk typically used?

Answer 16

cohort studies

Question 17

when is the odds ration typically used?

Answer 17

case-control studies

Question 18

when should odds ratios be avoided?

Answer 18

if a disease is common, odds ratio will overestimate the risk

Question 19

when to use chi squared test for trend?

Answer 19

when there is one nominal categorical variable and one ordinal categorical variable with at least 3 levels.

Question 20

Chi-squared test for trend hypothesis

Answer 20

H0: there is no linear trend in the relationship between variable x and variable y

Question 21

what are the degrees of freedom in a Chi-squared test for trend?

Answer 21

always 1

Question 22

how to interpret Chi-squared test for trend?

Answer 22

look at counts in the table to determine the nature of relationship; test only tells you if a relationship is present

Question 23

McNemar’s null hypothesis

Answer 23

H0 = there is no difference in the number of individuals between the first and second occasions

Question 24

what are the McNemar’s test assumptions?

Answer 24

1. data must be paired
2. response variable must be binary
3. number of discordant pairs must be large, ideally b + c > 10
4. each observation should correspond to a unique individual or a matched pair

Question 25

what are discordant pairs in McNemar's?

Answer 25

> participants whose outcome changes between readings > outcome has worsened or gotten better > these are the observed frequencies

Question 26

expected frequencies in McNemar's?

Answer 26

(b+c) / 2 > where b and c are discordant pairs

Question 27

When is Cochran Mantel Haenszel Test used (data type)?

Answer 27

used to assess the association between two categorical variables while controlling for a third stratifying variable

Question 28

What is a confounding variable?

Answer 28

> a third variable in a study examining a potential cause-and-effect relationship For a variable to be a confounder, it must: 1) associated with the exposure (i.e., it differs between the exposed and unexposed groups). 2) Be associated with the outcome (i.e., it influences the outcome). 3) Not be in the causal pathway.

Question 29

what is a stratifying variable?

Answer 29

a variable used to divide data into subgroups (strata) to analyse the exposure-outcome relationship within each subgroup.

Question 30

What is the CMH test null hypothesis?

Answer 30

H0 = There is no association between variable1 and variable2 after adjusting for the stratifying variable, variable3.

Question 31

What are the CMH test assumptions?

Answer 31

>applied to a series of 2x2 tables > each expected cell count should be at least 5 > observations are independent > if 0 values are present at all, it is not reliable

Question 32

CMH steps

Answer 32

1. Run a chi- squared/Fisher's to first determine if there is a relationship between the 2 variables (state hypotheses) 2. Then carry out CMH to determine if there is lurking or stratifying variable (restate hypotheses) 3. in conclusion, state whether stratifying variable is lurking or not

Question 33

When do you use a Kappa test?

Answer 33

a statistical measure used to assess the agreement between two raters (or observers) when they are classifying items into categories.

Question 34

what makes a Kappa test different?

Answer 34

Its not a hypothesis test; there are no p-values

Question 35

Cohen's Kappa interpretation

Answer 35

>maximum value of 1 when agreement is perfect a) <0.2 = poor b) 0.2-0.4 = fair c) 0.4-0.6 = moderate d) 0.6-0.8 = good e) 0.8-1 = very good > give interpretation for both weighted and unweighted Kappa.

Question 36

How to know if Cohen's Kappa is incorrect?

Answer 36

> weighted Kappa should be larger than unweighted Kappa because it accounts for the degree of disagreement rather than treating all disagreements equally

Question 37

What must be done in R to calculate Kappa?

Answer 37

> weighted Kappa matrix must be created > has 1s on the main diagonal >top row runs from 1 > 0

Question 38

Calculating Kappa matrix values

Answer 38

>take number of categories (columns) and minus 1 >take reciprocal of this ( 4 = 1/4) >move in increments of 1/4 between 1 and 0

Question 39

what are the limitations of Kappa test?

Answer 39

1) prevalence problem = if one category is much more common, kappa can be misleadingly low even if agreement is high 2) more categories lower kappa, fewer categories inflate kappa

Question 40

How do you combine factor levels in a table? (for Fisher's)

Answer 40

1) data$variable <- as.factor 2) levels(data$variable) 3) data$variable <- factor(data$variable, labels = c())

Question 41

How do you reorder factor levels?

Answer 41

1) data$variable <- as.factor 2) levels(data$variable) 3) data$variable <- factor(data$variable, levels = c())

Question 42

How to interpret relative risk of less than 1?

Answer 42

1) i.e rr = 0.6 2) exposed individuals have 40% less risk of event 3) if rr < 1, 1 - rr for interpretation

Question 43

How to make a table in r with no dataset

Answer 43

matrix( c( n, n, n, n,), ncol=2, byrow = true)

Question 44

what is the key difference between fisher's exact and chi squared?

Answer 44

Fisher’s Exact Test is conservative, sometimes leading to higher p-values than chi-squared

Question 45

When is the odds ratio (OR) preferred over relative risk (RR)?

Answer 45

when the event is rare