PAS 3 Flashcards
What does taking the mean of multiple observations do to uncertainty?
Decreases but does not eliminate uncertainty.
What is the percentage CI of around 1 sem around the mean
68%, meaning there is a 68% chance that the real answer is within that interval.
What does a region of +2 sem around the mean mean?
It is a 95% CI
What is an issue of sem with small numbers of observations?
Underestimates MOE
What do you need to do for triplicate observations sem?
what needs to be done to sem for triplicate observations
You need to double the sem error bars to get a 68% CI
What are the 2 types of experiments and what do they mean?
What is paired data
Have independent measures
Have repeated measures: where you have paired data, multiple repeated measures. Paired data typically arises when you make measurements on the same subject (before and after treatment)
What is effect size
Difference between means of two groups
How can uncertainty be estimated?
Uncertainty can be estimated roughly by looking at sem error bars. Or more exactly by calculating a 95% CI for the difference in the means
What is wrong with the old statistics?
a. Based on a wrong conceptual model
i. There is an effect (alternative hypothesis is true) or no effect (alternative hypothesis is true). Ignores effect size and biological significance.
b. Encourages definite decisions (accept or reject null hypothesis) based on inadequate data.
i. P ≤ 0.05 is only weak evidence for a real effect. P > 0.05 is little or no evidence against a real effect
c. Uses the semantically misleading term “statistically significant”
i. It really means the data is statistically indicative
ii. It does not tell you about the biological significance of the real effect
10. For paired data, why measure each subject twice, before and after?
This eliminates the uncertainty due to variation between individual subjects
How do you estimate effect size for paired data?
- calculate the effect size for each individual
- take the mean
How do you estimate uncertainty for effect size
Calculate the 95% CI for the effect using a paired t-test
Do not try comparing sem error bars
What is bonferroni correction?
Multiply the p-value by the number of independent tests. if it is still less than 0.05, then it may be significant.
Why perform conservative test
helps to avoid Type 1 errors (false positives) but may cause a lot of Type 2 errors (false negatives)
What does Post hoc mean
Afterwards tests
Example of a post hoc test
Tukeys post hoc test. This compares 2 groups at a time. Gives an individual confidence interval and p value for the comparison. It includes a multiple testing correction.
What is ANOVA
It is a one way analysis of variance.
• Looks at all of the data in all of the groups together
• Looks at the overall variation (variance) within the groups
• This measures the overall variability of the data
• Then looks at the overall variation (variance) between the groups
• Is the variability within the groups sufficient to explain the variation between groups?
• Technically, what is the probability of getting that large a variance between groups, if there is no real effect, so assuming only random variation (null hypothesis is true)?
• Looks at all of the data in all of the groups together
• Looks at the overall variation (variance) within the groups
• This measures the overall variability of the data
• Then looks at the overall variation (variance) between the groups
• Is the variability within the groups sufficient to explain the variation between groups?
• Technically, what is the probability of getting that large a variance between groups, if there is no real effect, so assuming only random variation (null hypothesis is true)?
Why are send not used for small number of observations
Underestimated the uncertainty
What are the types of experiments and give examples
Independent measures eg patients vs control or transgenic animals vs control animals
Repeated measures, paired data or multiple repeated measures
Why should you not compare sem error bars for compared data
Idk
What is p hacking
Crossing out or not involving certain data.
Eg is p=0.052, close to 0.05, so they change experiment condtions to get p<0.05
Changing the measurement to get p as smaller than or equal to 0.05
or to make all sorts of different comparisons and tests, and if one of them comes out significant, you publish that one and do not publish the others.
Publishing significant data and leaving out unsignificant ones. This is scientific fraud
Structure of spliced rna called ?
Lariat
In an independent groups experiment
the effect size is the best estimate of the true answer. The uncertainty (MOE) can be estimated from the sem error bars. This gives a roughly 80% CI, (assuming at least 10 observations in each group and a rough approximate distribution). A better approach is to use a t-test to calculate the exact 95% CI, and to use half the 95% CI instead of sem on the error bars on the graph
For paired data
The sem error bars for each group are not a good indicator. Calculate the difference for each pair of data points. Then plot the mean and 95% CI for these results.
Multiple groups and multiple tests
Increased likelihood of type 1 errors.
Important to show all tests and comparisons
if there are a large number of comparisons.
a) use a multiple ttesting correction or
b) use anova followed by a post hoc test.