Logistic Regression Basics
A system for naming and organizing independent variables and their coefficients in a regression model
EVW Model
A useful way to see if the relationship between your exposure and outcome differs within strata of another variable
Interaction assessment
A way to describe a model where all lower order components of the variable must be included in the initial model
Hierarchically Well-Formulated (HWF)
Type of logistic regression that is appropriate when the outcome has >2 levels and the levels have an inherent order, and when requisite assumptions are met
Ordinal logistic regression
This type of regression is appropriate when the outcome is dichotomous, not rare, and when the study design is cross-sectional or a cohort study
Log binomial regression
Model to find a valid estimate between an exposure and a dichotomous health outcome
Logistic regression
When using regression, this is included in a model to assess interaction
Product term
If you were to follow the recommended strategy, this is the step you would perform after variable specification and collinearity assessment
Interaction assessment
This statistical test is used to evaluate whether the proportional odds assumption is met, and whether ordinal logistic regression can be performed
Score test
A method of confounding control whereby comparison group members are intentionally selected so that they are similar to the index group with respect to the confounder
Matching
Calculation to transform modeling probabilities to odds ratios
ln(odds) or ln(p/1-p)
Variables which represent nominal independent variables and cannot be separated in a model
Indicator (dummy) variables
During the confounding assessment, this is what you would conclude if the OR produced when dropping a potential confounder is within 10% of the OR produced from the gold standard model
No evidence of confounding
For this type of logistic regression, the analyst does not have to choose the reference category for the outcome
Ordinal logistic regression
These are the two measures of association that are produced when performing logistic regression for a risk-type cohort study and a cross-sectional study
ROR and POR
The effect of a one unit increase in the exposure variable on the outcome, controlling for confounders
e^(B1)
Used in a contrast statement to measure interaction with a nominal variable
Chunk test
Your exposure is a 4-level nominal variable. You decide to create 4 dummy variables and include them all in your model. This is the reason why your model does not run
Collinearity
You have a single, dichotomous exposure. Your outcome is nominal and has 3 levels. You also have an effect modifier with 4 levels. This is the number of ORs that can be estimated from polytomous logistic regression
(G-1)*(levels of EM) = (3-1)*(4) = 8
This is the primary advantage of mathematical modeling as opposed to Mantel Haenszel methods in the analysis of matched data
Can control of matched and unmatched factors
(1) Number of models required to conduct and (2) number of parameters that can be tested by the likelihood ratio test (LRT) and Wald test.
(1) LRT: 2; Wald: 1
(2) LRT: no limit; Wald: 1
The test statistic given when using a contrast statement in proc logistic
Wald chi-square
The process of backwards elimination
Evaluate the product terms in the model by running the model and assessing if any product terms have p > alpha. If so, remove the least significant term and run the model again. Repeat this process until all parameter terms are either significant or have been dropped from the model.
The degrees of freedom for the likelihood ratio test to determine the significance of a single exposure in polytomous logistic regression
1 less than the levels of the outcome (G-1)
The EOR estimated from a case-control study equals the ROR estimated from a cohort study when controls are sampled in this particular way
"Row 2 sampling" (truly free of disease)