Data Analysis Workshop 2

Question 1

What is a t-test?

List the assumptions of a t-test.

Answer 1

• A t-test determines if the means of two datasets are significantly different from each other.

Assumptions include:

1 - The data is normally distributed.

2 - The populations being compared have the same variance (however the t-test is quite robust to unequal variance).

3 - The samples are independent from one another.

Question 2

What is parametric data?

Answer 2

Parametric data is data that is normally distributed.

Question 3

What is a one-way analysis of variance (ANOVA) test?

Why use an ANOVA test rather than just many individual t-tests?

What are the assumptions of an ANOVA test?

Answer 3

• An ANOVA test determines if the means of three or more are significantly different from each other.
• If an ANOVA test returns a statistically significant result, this indicates that at least two group means are significantly different from each other.
• An ANOVA test cannot tell you which specific groups are statistically significant from each other, therefore a post hoc test is required.
• An ANOVA test is used before post hoc tests (e.g. a t-test) because by doing one statistical test rather than many individual tests, the total error (p value) combined from all tests is lower.
• The assumptions of an ANOVA test are the same as for a t-test.

Question 4

What is a Bonferroni correction?

When does this not need to be done?

Answer 4

• A Bonferroni is used to reduce the acceptance p value (normally 0.05) of the false positive to account for the increase in error that results from using many statistical tests.
• If multiple tests are done with hypothetical data, a Bonferroni correction does not need to be done.

Question 5

How can variance homogeneity be tested?

Answer 5

Homogeneity of variance can be tested using a Levene’s test.

Question 6

List 3 ways of measuring the association between two datasets that are non parametric.

Answer 6

Association between non parametric datasets can be measured by:

1 - Using a nonparametric test such, as a Kruskal-Wallis H test.

2 - Transforming the data to achieve normal distribution.

3 - Using a parametric test that tolerates nonparametric data well, such as a t-test or ANOVA test.

Question 7

List 2 ways of measuring the association between two datasets that do not have homogenous variance.

Answer 7

Association between datasets with heterogeneous variance can be measured by:

1 - Using a nonparametric test, such as a Kruskal-Wallis H test.

2 - Using a Welch or Brown and Forsythe test.

Question 8

How can association be measured between datasets that are not independent from one another?

Answer 8

If there is a lack of independence in data, there is nothing that can be done to determine association, and requires a redesign of the study.

Question 9

What is a correction analysis used for?

What is a regression analysis used for?

What is a linear regression analysis used for?

Answer 9

• Correction and regression analyses is used to measure association between two continuous variables.
• The difference between the two is that a regression analysis fits a curve and gives you a formula that describes the relationship.
• A linear regression models the relationship between two variables by fitting a linear equation to the data.

Question 10

What is a receiver operating characteristic (ROC) curve?

Answer 10

• An ROC curve is used to show graphically the relationship between clinical sensitivity and specificity for every possible cut-off for a test or a combination of tests.
• The curve is used to choose the most appropriate cut-off for a test, which has the highest true positive rate together with the lowest false positive rate.
• The area under an ROC curve is a measure of the usefulness of a test in general.