Biostats E1

Statistical tests, their assumptions and more!

3 types of Variables:

1. fixed vs random

2. discrete vs continuous

3. numerical vs categorical

4. independent vs dependent

Errors of data organization:

1. data is not versioned

2. Shorthand dictionary is missing

3. data has not been backed up

4. headers are wordy with lost of spaces and no clear pathway

Type 1 error:

rejecting the null when it should be accepted

type 2 error:

accepting the null when it should be rejected

When to use a chi-squared test:

comparing the goodness of fit between two groups of categorical data

When to use a t-test:

When comparing the means of numerical data

4 types of t-tests:

1. paired

2. unpaired

3. one tailed

4. two tailed

What does an F statistic mean:

gives the ratio of variances between two groups

When to use a Cohen’s d test:

to determine what strength of influence subgroups have on the results (0-.2 small, 0.5 medium, .8-1.4 large)

When to use a correlation test:

determines the direction and influence between two variables when it is unknown, anything greater than 0 indicates correlation

When to use a linear regression:

use to determine causation

When to use a regression model:

when you know that the predictor variable influences the response variable

assumptions to a linear regression:

1. observations are taken from a random sample

2. individual observations are independent from each other

3. predictor variable is not a random variable

4. for each X value, the corresponding Y comes from a normal distribution

5. variance is constant and does not depend on H (homoscedastic)

When to use a Wilcoxon Signed Rank test:

when data is non-normally distributed and you want to compare the means (nonparametric version of t-test)

r means:

correlation coefficient, indicates the direction and magnitude of correlation (-1 to 1)

assumptions of a correlation test:

1. both variables are continuous

2. a linear relationship between the variables exists

3. no outliers are present

4. both variables are normally distributed

What test does a Spearman’s correlation test relate to:

Nonparametric version of the correlation test

Possible transformations:

1. log

2. square root

3. box-cox

4. arcsin

5. power

6. exponential

7. reciprocal

When to use nonlinear models:

when transformations cannot alter data enough so a parametric test can be used

two nonlinear models:

Michaelis-Menten and quadratic tests

ANOVA assumptions:

1. Normality

2. Homogeneity

3. Additivity

Regression Fallacy

As more samples from an individual are taken, they will regress towards the mean, however this is not due to treatment

What a Poisson distribution is:

Describes the number of successes in blocks of space and time

Poisson Distribution assumptions:

1. Successes are independent of one another

2. Successes must occur with equal probability

When to use Poisson Distribution:

to determine whether events happen randomly in space and time

When to use Log Regression:

when responses are binary so a linear regression cannot be used

Appearances of linear vs. logistic regressions:

linear= straight, logistic= curved