Measures of Evidence in Therapeutics Flashcards
What is data?
Factual information (ie. measurements or stats) used as a basis for reasoning, discussion, or calculation
What is an inference?
Coming to a conclusion given info or premises by any acceptable form of reasoning
What is evidence?
Material that proves or disproves something; grounds for belief; proof
How would you differentiate data and evidence?
Data is a basis for reasoning, but it must be combined with a process of inference before it can be called evidence
Deductive inference
Starting from specific premises and forming a general conclusion
Ex. Flu causes the symptoms of aches, fever, headaches, and fatigue
Inductive inference
Using general premises to form a specific conclusion
Ex. Aches, fever, headaches, and fatigue may be the result of the flu
Bayes Theorem
Describes the probability of an event, based on prior probability that might be related to the event
The distribution of probabilities for each possible outcome forms the probability distribution (which should sum to 1)
Credible interval as measure of evidence
Acts as a “likelihood” function, but this function doesn’t integrate to 1, therefore, it cannot provide the probability of any result. It is a measure of evidence in Bayesian statistics because it relies on its assumptions
An example of a credible interval is the 95% credible interval, which assumes that there is a 95% probability that the true parameter value lies within that interval
(look at slides 17-20)
What is the main difference between Bayes theorem and hypothesis testing?
Hypothesis testing avoids the “prior” part subjective. It frames a research question in the form of 2 hypotheses, then performing a test using observed data. Given the data, one hypothesis will be rejected and the other will be accepted based solely on the observed data
What are the 2 risks of Neyman’s and Pearson’s proposed “behaviour”?
- Rejecting the null hypothesis when it is true (type I error rate, α)
- Not rejecting the null hypothesis when it is false (type II error rate, β)
Confidence interval as measure of evidence
Confidence interval is a “classical” or “frequentist” statistic, meaning the parameters are considered fixed, but unknown
In a 95% CI, we have to assume that HR is fixed, meaning out of 100 RCTs (randomly controlled trials), 95 of the samples will contain the true underlying HR. We don’t know which 95 are correct
Likelihood ratio as measure of evidence
Comparison of how well 2 hypotheses (effect sizes) represent the data. The effect size that better predicts the data has more evidential support
LR offers a compromise between Bayesian and frequentist perspectives, such that it is mathematically related to the data part of Bayes theorem, but it doesn’t include the prior part
P-value as a measure of evidence
It’s the probability of obtaining a result that is equal to or more extreme than what was observed when it is assumed the null hypothesis is true
It is NOT used for inference
Null hypothesis
No relationship exists between two sets of data or variables being analyzed
What are 3 reasons as to why P value is not usable for inductive inference?
- P is a purely negative and non-relative measure. It tries to say it is against the null
- But P doesn’t even say what it’s against because it’s calculated under the assumption that the null is true, therefore, it can’t be a direct measure of evidence that the null is false
- A large effect in a small trial and a small effect in a large trail can have the same P value. This is an issue if effect size is important for accepting a hypothesis of “no effect”
