Evidence Based Practice Week 11 – Causation and Correlation

Correlation Basics

• Correlation signifies an association between variables, measurements, or outcomes, where a change in one corresponds to a change in the others.
• Correlations range in strength from weak to strong, with moderate correlations in between.
• Correlation tests yield an "R" value between -1.0 and +1.0, where 0 indicates no correlation.
• A negative R value (e.g., -0.1) indicates an inverse relationship: as one variable increases, the other decreases.
• A positive R value (e.g., +0.1) indicates a direct relationship: as one variable increases, the other also increases.
• The absolute value of R indicates the strength of the correlation.
• A strong correlation is typically indicated by an R value greater than 0.7.
• An R value of 1.0 represents a perfect correlation, with variables increasing proportionally in a linear fashion.
• A statistically significant correlation requires both a notable R value and a p-value less than 0.05.
• A strong R value (e.g. 0.95) is insignificant if the p-value is greater than 0.05, potentially due to a small sample size.
• Caution is needed when interpreting weak correlations as significant, even with a p-value less than 0.05.
• Pearson’s product-moment correlation is suitable for continuous, normally distributed data.
• Spearman’s rho is used for non-normally distributed or ordinal data.
• Both tests result in an R value.

Causation Basics

• Causation implies a cause-and-effect relationship between variables, where a change in one directly affects the other.
• Correlation does not equal causation.
• Confounding factors influence both the intervention and the outcome, potentially creating the illusion of a causal relationship when one does not exist.
• Mediators explain the cause-and-effect relationship during an intervention.
• Responsiveness measures the ability of an outcome to detect meaningful changes over time due to an intervention.
• Extrapolation involves applying findings from one group to another, which may be inappropriate depending on physiological and biological differences.
• Prediction accuracy relies on the strength of the relationship between the predictor measurement and the outcome variable.
• Multiple predictors can lead to more accurate predictions of an outcome.

Confounding Explained

• Confounders are variables that influence the relationship between an exposure and an outcome.
• Colliders are caused by both the exposure and the outcome.
• Mediators are caused by the exposure and, in turn, cause the outcome.
• Exposure examples include risk factors, prognostic factors, predictors, independent variables, influencing factors, associated factors, and treatment types.
• Randomization balances confounders between groups, reducing bias risk.
• Statistical adjustments can control for confounders using methods like multivariable regression, matching, propensity scoring, weighting, and instrumental variables.
• Reducing confounding risk involves identifying potential confounders and adjusting for them.
• Directed acyclic graphs help researchers visualize their theories and identify confounders for adjustment.
• A causal factor increases or decreases the likelihood of an outcome, while not being the only cause.

Assessing Confounding Risk

• Determine if the analysis adjusts for confounders.
• Analyze how confounders were selected.
• Be aware of simple statistical processes which may include variables that are not confounders.
• The ommission of significant confounders like sex, age, and baseline severity could signify researcher error.
• Be suspicious if a variable that influences both the exposure and outcome has not been accounted for.
• Ensure that any theoretical models or directed acyclic graphs used include all relevant confounders.