p and CI
p-value
Assume that we have a normally distributed population of males and females with a mean height of 168.5 cms and 157.8 cms respectively.
We can see that the mean difference in the height is 10.7.
If I measure the entire population in the world, will I find that the mean height for each sex is different? or would I find that difference between the mean height is zero?
How to answer this question?
Frequentest method
Assume a hypothetical world with no difference in the height between the sexes.Do repeated experiments and find out what proportion of times would I obtain a difference between the observed means (ø) in that hypothetical world which is as great as or greater than difference I observed in my real experiment. This proportion of time is abbreviated as p-value
In other words p-value=P(ø/H),i.e., given the null (H), what is the probability of data(ø)
but p-value is an answer to a question we didn’t intend to ask. So what to do with it?
Traditional Answers
Fisher
Treat it as a measure of strength of evidence.
This is not objectively satisfactory because, p-value=P(ø/H)≠P(H/ø)
Neyman-Pearson decision framework
In each experiment, if you need to choose between the Null and the alternative, then there is possibility for two types of errors
False Positive: Choosing the alternative when Null is true
False Negative: Choosing the Null when the alternative is true
Control the rate at which you commit decision errors of false positive type by applying a strict procedure to each experiment:
The researcher has to make a decision. In research, convention is to choose the alternative hypothesis if the test result is outside the 95% mark in the normal curve. The remaining 5% is called œ. The boundary is approximately 1.96z, ie, 1.96 standard deviations away from the mean.
A Z-score is a numerical measurement that describes a value's relationship to the mean of a group of values.
Z-score is measured in terms of standard deviations from the mean. If a Z-score is 0, it indicates that the data point’s score is identical to the mean score. A Z-score of 1.0 would indicate a value that is one standard deviation from the mean. Z-scores may be positive or negative, with a positive value indicating the score is above the mean and a negative score indicating it is below the mean.
œ is the zone of rejection of null hypothesis and z the boundary between the acceptance and rejection zones of null hypothesis.
If we refer the statistical charts, p value for a z score of ≤1.96 is ≤0.05
Decision framework
So the decision framework is to
Compute p-value
Fail to reject the Null if p-value is > α
Reject the Null if p-value is <= α
Typically α is set at 0.05
Bayesian approach
Bayesian approach is more humane and direct. The researcher ‘should’ have prior believes about the probability of null and alternate hypotheses. She then construct a posterior probability of null from the joint probability of finding data for null p(d,H0) and the marginal probability of data p(d)
p(H0|d) = p(d,H0)/p(d) = p(d|H0)* p(H0)/p(d)
Confidence Interval
Under repeated sampling in hypothetical world, 95% of confidence intervals constructed the same way will include the true value of the mean difference.
It ranges between a fixed value x and y on either side of zero,i.e., x-0-y.
0-x is the lower confidence limit and 0+y is the upper confidence limit.
Confidence limit
This sets the boundary of confidence interval and it is generally taken as 95%.
Confidence interval is generally considered as more important than p value because it gives
Statistical significance
Clinical significance
Precision