stats

Question 1

Mean greater than median, which skew?

Answer 1

positive

Question 2

Median greater than mean, which skew?

Answer 2

Negative

Question 3

SD formula?

Answer 3

Square root: Sum of (x1-xmean)squared /n-1

Question 4

…% data is within 1SD of the mean?

…% data is within 2SD of the mean?

Answer 4

68% (32 outside)

95% (5 outside)

Question 5

Equation for the Z value?

Answer 5

Z=x1-xmean/SD

Question 6

Confidence Interval equation?

Answer 6

Mean + or - (1.96x SE)

Question 7

SE equation?

Answer 7

SD/ square root of n

Question 8

SE of the difference between two means equation?

Answer 8

SE= square root [(SD^2/n1) + (SD^2/n2)]

Question 9

SE of proportions equation?

Answer 9

SE= square root [p(1-p)/n ]

where P is the risk of outcome

Question 10

SE of difference between proportions equation?

Answer 10

SE= square root [p(1-p)/n + p(1-p)/n2]

where P is the risk of outcome

Question 11

Statistically significant if..?

Answer 11

P value less than 0.05 (or 0.01 if that’s the cut off)

CI doesn’t include zero (or the value specified in the null hypothesis e.g. if its a risk ratio not 1)

Question 12

false negative rate=

false positive rate=

Answer 12

false negative rate(type II)=1- sensitivity (Neg and seN)

false positive rate (type 1)=1-specificity (Positive and sPec)

Question 13

Normal or reference range?

Answer 13

the 95% CI range

Question 14

Standard deviation vs SE use?

Answer 14

SD- describes variability of the data/ spread

SE Measure if the precision of an estimate, used to form a CI, which can infer significance.

Question 15

If binary independent variable and continuous dependent variable?

Answer 15

T-test, paired or unpaired according to whether it is the same individual or different

Question 16

If paired binary independent variable and continuous dependent variable but not normally distributed?

Answer 16

Wilcoxon matched pairs test

Question 17

If paired samples but the data is ordinal dependent variable, and binary independent variable?

Answer 17

Wilcoxon matched pairs test/signed rank

Question 18

If paired samples but both data is binary?

Answer 18

Mcnemars test

Question 19

If unpaired binary independent variable and continuous dependent variable but not normally distributed?

Answer 19

Mann- whitney U test

Question 20

If multiple unpaired categorical independent variable and continuous dependent variable?

Answer 20

ANOVA

Question 21

Two unordered categorical variables (nominal or binary)?

Answer 21

Chi squared test

e.g. outcome or not is binary- makes a contingency table of rates etc

Question 22

Chi squared test requirements?

Answer 22

Unordered categorical data
Can form into a contingency table
80% expected n’s above 5
all expected counts over 1.

Question 23

If 3+ ordered categorical variables and a nominal/ binary independent outcome data?

Answer 23

Chi squared test for trend (e.g. continancy table lots of ordered exposures e.g. in dose response)

Question 24

If two continuous variables normally distributed?

Answer 24

Regression/pearsons correlation coefficient.

Question 25

If two continuous variables not normally distributed?

Answer 25

Spearmans correlation coefficient (or ordinal X variable0

Question 26

If continuous independent variable but binary dependent variable?

Answer 26

Logistic regression

Question 27

if more than 2 independent variables but continuous dependent?

Answer 27

multiple linear regression

Question 28

If pared sample- degrees of freedom?

Answer 28

n-1

Question 29

If pared t-test sample- degrees of freedom?

Answer 29

n-1

Question 30

if non paired t-test sample- degrees of freedom?

Answer 30

n-2

Question 31

Chi squared degrees of freedom?

Answer 31

number of categories-1.

Question 32

test options for continuous data?

Answer 32

Outcome continuous:paired t-tests, Wilcoxon (non paramtric) or independent t-tests and Mann-whitney if non parametric. ANOVA if multiple exposures. both: Regression/pearsons correlation. Spearmans rank if not normal. Multiple linear regression if multiple. Continuous exposure data: Logistic regression (binary outcome) or spearmans rank (ordinal)

Question 33

Test options for categorical data?

Answer 33

Two unordered: Chi squared Ordered outcome: chi squared test for trend. Paired proportions: McNemar

Question 34

Multiple linear regression equation?

Answer 34

Y=b0 (intercept) + (b1x1) +(b2+ x2)…. where b1=the regression of the first variable and X1 is the subjects exposure i.e. (Male=0 and female=1) with r=0.2 if female = (1x0.2)

Question 35

R squared is..

Answer 35

The variation of the dependent variable that is explained by the independent variable. 1-r2 = the unexplained variation E.g. r2=0.66 then 66% of variance is explained

Question 36

Regression assumptions?

Answer 36

X does not need to be normally distributed, but the residuals (deviations from the regression line) need to be normally distributed for the X values. Variance the same for X and Y Linear relationship

Question 37

specificity is=1- | Sensitivity is=1-

Answer 37

Sensitivity= 1-proportion of false negatives (HAVE DISEASE) | specificity=1-proportion of false positives (WITHOUT THE DISEASE)

Question 38

Likelihood ratio positive equation? WHat is it?

Answer 38

sensitivity/1-spec | Likelihood of positive result with the disease vs without

Question 39

Likelihood ratio negative equation? WHat is it?

Answer 39

1-sens/spec | Likelihood of negative result with the disease vs without

Question 40

standardised effect size calculation?

Answer 40

Mean of treatment- mean of control/ SD, where 1 means 1SD away

Question 41

What determines sample size? (4)

Answer 41

1. Target anticipated effect size/ difference 2. The variability/ SD of the outcome data 3. Power (typically 80%) 4. The significance level (0.05)

Question 42

sample size increases with? (4)

Answer 42

Binary outcomes, large SD, lower signifincance level, higher power.

Question 43

what is needed for a sample size calculation for a normally distributed RCT? (4)

Answer 43

1. Type 1 error 2. Type 11 error 3. SD of outcome 4. Size of difference to detect

Question 44

what is needed for a sample size calculation for a binary RCT? (4)

Answer 44

1. type 1 error 2. Type 2 error 3. anticipated % with the outcome on new and standard treatment (read difference off table) 4. size of difference want to be able to detect. (SD not needed)

Question 45

what is needed for a sample size calculation for estimating the prevalence of a disease in population (survey)? (4)

Answer 45

1. Estimate of the disease prevalence 2. The confidence level (e.g. 5% either side) 3. How precise the estimate should be. 4. Estimated response to survey

Question 46

If it says the trials results are unadjusted which test to use?

Answer 46

Chi squared, if is adjusted logistic regression

Question 47

what is needed for a sample size calculation for a cluster RCT? (4)

Answer 47

1. Average no. of patients per cluster 2. Risk of type I error 3. Size of difference to detect 4. Intracluster coefficient (not correlation coefficient)

Question 48

ROC curve plot?

Answer 48

1-spec on X axis, sensitivity on y

Question 49

LR (+) if high or low use?

Answer 49

If low useful for ruling out a disease, if high useful for ruling in a disease

Question 50

T test T statistic equation?

Answer 50

Paired: mean difference/SE=t (n-1) Unpaired: Mean X1-Mean X2/SE=t (n-2)

Question 51

CI for relaive risks?

Answer 51

SE= square root [1/a-1/a+c + 1/b -1/b+d]