Stats

Question 1

Null hypothesis (H0)

Answer 1

No difference between the two groups (default hypothesis)

Question 2

Alpha (type I) error

Answer 2

Incorrect rejection of the null hypothesis - conclude there is a difference when there is no true difference

Question 3

Beta (type II) error

Answer 3

Incorrect acceptance of the null hypothesis - conclude there is no difference when there is

Question 4

Standard deviation

Answer 4

Measure of the variability of the data from the mean, equal to the square root of the variance

Question 5

Standard error

Answer 5

Measures the variability of means when many similar samples were taken from the population of possible measurements - how close the sample mean is likely to be to the true population mean

Question 6

Power

Answer 6

The ability to detect a significant result where a true difference exists (the probability of rejecting the null hypothesis when the alternative hypothesis is true)

Question 7

Accuracy

Answer 7

How close a value is to the true value

Question 8

Precision

Answer 8

How consistent results are when measurements are repeated

Question 9

Validity

Answer 9

Suitability of the experimental method to address the aim of the experiment

Question 10

Internal validity

Answer 10

If outcome is related to aim

Question 11

External validity

Answer 11

If results are applicable to the wider population

Question 12

Odds ratio

Answer 12

How much more likely one group is to be in an outcome group than another

Question 13

What is the alpha (type I) error also known as?

Answer 13

Specificity

Question 14

What are alpha (type I) errors directly related to (statistically)?

Answer 14

P value

Question 15

What are beta (type II) errors also known as?

Answer 15

Sensitivity

Question 16

What is the equation for calculating power (in simple terms)?

Answer 16

1-beta error

Question 17

What power value do we typically want an experiment to have?

Answer 17

Between 0.8 and 0.9

Question 18

What does power relate to (statistically)?

Answer 18

Sample size and effect size. If sample size is large enough, even a tiny effect size may be statistically significant (but not necessarily clinically or functionally significant)

Question 19

When may precision and accuracy become uncoupled?

Answer 19

Systematic error

Question 20

When is mean vs median usually used?

Answer 20

Mean - when data are normally distributed with no major outliers

Median - when data are skewed or non-normally distributed

Question 21

If the median of data is used, what should be used to measure spread?

Answer 21

Interquartile range

Question 22

If the mean is used, what should be used to measure spread?

Answer 22

Standard deviation - spread of direct data

Standard error - where the real mean may be based on the population

X% confidence intervals - the range containing the population mean X% of the time based on the data collected

Question 23

How can we test for normality?

Answer 23

Visually of q-q plots

Using tests such as the Shapiro-Wilk test

Question 24

If data are normally distributed, what type of test is used?

Answer 24

Parametric

Question 25

If data are not normally distributed, what type of test is used?

Answer 25

Non-parametric tests

Question 26

What do non-parametric tests typically do?

Answer 26

Rank data and analyse whether a difference exists in the average rank between groups

Question 27

Why is ANOVA preferable to multiple t tests?

Answer 27

Reduced risk of type I errors which would occur due to multiple comparisons

Question 28

What does ANOVA tell you? What does this mean?

Answer 28

That a difference exists between any of the groups tested, but not where. One must then use post-hoc tests to compare means and identify where the difference is (these will include corrections)

Question 29

What does the Tukey test do?

Answer 29

Compares

Question 30

What does the Tukey test do?

Answer 30

Compares the differences in means between all groups

Question 31

What does the Dunnett correction do? When may this be used?

Answer 31

Compares the mean of each group to a control value - could use when samples are taken periodically and you want to test when they become significant (known boundary)

Question 32

What does the Bonferroni correction do? How?

Answer 32

It adjusts the p value for multiple testing - p is changed to be equal to alpha divided by the number of tests performed

Question 33

What are the 5 main assumptions of T tests?

Answer 33

Data are normal Groups are independent (unpaired)/ samples are independent (paired) Equal variance Measurements are continuous No significant outliers

Question 34

What are the 4 main assumptions in ANOVAs?

Answer 34

Data are normal Variance is equal Samples are independent Observations within each sample are random

Question 35

What are the 4 assumptions of Pearson's test?

Answer 35

Continuous variables Paired values for each observation No outliers for either variable Data are linear

Question 36

What information is required to calculate the number of replicates required to detect a known true difference?

Answer 36

The variation and size of difference expected, the accepted alpha/beta threshold - use past literature

Question 37

When shouldn't non-parametric tests be used?

Answer 37

When n is very small i.e. below 6

Question 38

What is the median?

Answer 38

Value in the middle when listing them all. ## Footnote Great for non-normally distributed data.

Question 39

What is the mean?

Answer 39

Average of all values. ## Footnote Skewered by outliers.

Question 40

What is a hazard ratio?

Answer 40

Treatment hazard rate divided by placebo hazard rate. ## Footnote E.g. 10% in treatment die, 20% in control die -> hazard ratio = 0.5 '50% decrease in deaths'. E.g. Hazard ratio = 0.64 -> 36% decrease in deaths in treatment compared to control.

Question 41

What is a confidence interval?

Answer 41

We can be 95% sure that the true number lies within that value.

Question 42

What are the two types of significance?

Answer 42

Statistical and clinical.

Question 43

What is statistical significance?

Answer 43

P value is less than 0.05. ## Footnote Note a given hazard ratio has a 0.005 p value then there is a 0.5% chance of this hazard ratio occurring despite no clinical difference.

Question 44

What is clinical significance?

Answer 44

Depends on the application and effect size. ## Footnote A statistically significant hazard ratio of 0.98 is only a 2% decrease in death. In clinic this could or could not be significant depending on how many people are affected. Another example: 1 mmHg reduction in BP with antihypertensive (that might even have side effects) is probably not significant in clinic.

Question 45

What are parametric tests?

Answer 45

Deal with normally distributed data. ## Footnote More likely to get a statistically significant result, usually higher power.

Question 46

What are non-parametric tests?

Answer 46

Deal with non-normal data. ## Footnote Either do a t-test to show difference between two groups (that is a parametric test?) Or transform data e.g. log transform, other exponents to make it normally distributed and use parametric tests.

Question 47

What is a technical replicate?

Answer 47

Taking the same reading multiple times on the same sample. ## Footnote E.g. BP of one individual three times to make sure measurement is correct.

Question 48

What is a biological replicate?

Answer 48

Same treatment to different samples. ## Footnote E.g. take BP of six different people.

Question 49

What are error bars?

Answer 49

Could be standard deviation, standard error, or 95% confidence interval. ## Footnote Standard error is commonly used over the standard deviation as its SD divided by sample size and thus makes the error bars look smaller.

Question 50

What is a test of multiple comparison, why is it important, and what is post-hoc testing?

Answer 50

Importance: p = 0.05 -> 1 in 20 tests will give a false positive. ## Footnote Thus do a test for multiple comparisons such as ANOVA. Post-hoc testing e.g. Bonferroni is then done to see differences between individual samples and a control or between all possible combinations - some tests have higher false positives other false negatives.

Question 51

What are the conditions to be able to run a t-test and what is the result?

Answer 51

Data normally distributed, equal variance in both groups of samples, data is continuous. ## Footnote Result: one p-value - whether or not there is a statistical significance between the two groups.

Question 52

What is the difference between one-way ANOVA and two-way ANOVA?

Answer 52

One way: multiple sample groups with only one variable e.g. effect of ramipril, amlodipine, both or placebo on BP. ## Footnote Two way: multiple sample groups with multiple variables e.g. ramipril, amlodipine, placebo effect on BP in old and young people or in high vs low doses.

Question 53

What is an ANOVA readout?

Answer 53

A single p value that tells you whether or not there is some difference between some of the data that you are testing.

Question 54

What do we do after an ANOVA?

Answer 54

Post hoc testing. ## Footnote E.g. run a t-test on each of the pairs with a correction such as Bonferroni.

Question 55

What is a Bonferroni correction?

Answer 55

If you’re doing five post-hoc tests you have to divide your p value by the number of tests. ## Footnote E.g. 5 -> p value of each post hoc test would have to be <0.01 for the result to be statistically significant.

Question 56

What is a power analysis?

Answer 56

Done before a study and takes into account: clinical effect expected, variance expected, statistical tests to be run, what acceptable false positive is. ## Footnote And computer answers you: which sample size is needed to make sure test will be significant and avoid false negatives.