Methods III Flashcards
In Hpyothesis testing, for which p-values is it likely to reject the null-hypothesis? (When p = P(observed effect | H0)
For (very) low values of p, it is a high confindence of the null-hypothesis to be rejected.
In Hpyothesis testing, for which p-values is it likely to not reject the null-hypothesis? (When p = P(observed effect | H0)
For (very) high values of p, it is a low confindence and statistically insignificant.
What are we testing in Hypothesis testing?
We are testing the likely hood of the measured difference between the means of a hypothesis and an actual sample.
If we reject the null-hypothesis, does this mean we accept the alternative hypothesis?
No
Do we reject the null-hypothesis if the p-value is above or under the significance level (alpha).
we reject when p-value is under (or equal to) alpha (p « a)
Do we fail to reject the null-hypothesis if the p-value is above or under the significance level (alpha).
When p-value is above the significance level a, we fail to reject the null-hypothesis.
In hypothesis testing, what is an Type II error?
When the truth is that H0 false, but hypothesis-test incorrectly not rejects H0
In hypothesis testing, what is an Type I error?
When the truth is that H0 is true, but hypothesis-test incorrectly rejects H0
In hypothesis testing, what would a false positive mean?
That the H0 actually is true but it still was rejected. (the attempt to reject the null-hypo was (falsely) positive) (type I error)
In hypothesis testing, what would a false negative mean?
That the H0 actually is false but it still was failed to reject it. (the attempt to reject the null-hypo was (falsely) negative) (type II error)
In hypothesis testing, what would a true negative mean?
That the H0 was true and hence it was failed to reject it. (the attempt to reject the null-hypo was (correctly) negative)
What is the chance of a Type II error in hypothesis testing?
This chance is equal to the area underneath the H1 curve that lays to opposite site of the signififcane level. We call this ß.
What is the chance of a Type I error in hypothesis testing?
This chance is equal to the significance level (alpha), because it represents the probability underneath the H0 curve that is smaller than alpha.
What is a t-test used for?
A t-test is used to define the chances of a curve in hypothesis testing.
What formula is used for t-tests? (you should be able to indentify what parameters have what influence)
t = sqrt(n) * (mean(X) - mu) / (sigma). Where mean(X) is the x axis and mu is the mean of an hypothesis. Sigma is the standard deviation.
Will an increase in sample size reduce or increase the Type II error of an t-test.
Reduce, as it sharpens the curve and hence reduces the area.
Will an decrease in the alternative hypothesis (mu) reduce or increase the Type II error of an t-test?
Increase, When the curves are moved more closely, the area (thus chance) of a type II error increases.
What change in variance will result in a higher type II error?
A higher variance, will allow for wider curves, and thus more area for a type II error.
What can we change to improve t-test for hypothesis testing?
As variance and mean are defined by the data itself; we can only gather more data in order to improve our t-test.
What is the power of a t-test?
1-P(type II) = 1 - ß
What does statistical testing relate to?
It only relates to random error, systematic error is completely ignored.
What is statistical validity?
Are our claims valid? Are the measured values really the practical truth?
What is statistical reliability?
Are we confident about our claims? Is it likely to obtain the same results in an equal study?
What is pseudoreplication?
Any two experimental units should be able to receive two different treatments, if they can’t they are probably sampling unit. Psuedoreplication: If we interpet this wrong in our observations, the false replication inflates the sample size, statistical power, etc.
In retrospect to internal validity, what is selection bias?
There are already (unknown) differences between groups.
In retrospect to internal validity, what is the threat of Repeated testing?
Subjects learn and remember between observations
In retrospect to internal validity, what is the threat of Matuartion and learning?
Subjects change during the study, affecting the response.
In retrospect to internal validity, what is the threat of Experimenter bias?
Researchers unknowingly affects subjects and their responses
In validity, what is regarded an non-experiment?
We only observe existent data; we do not control the treatments or patients.
In validity, what is regarded a quasi-experiment?
We can now control wich treatment is used, but that decision is not random: it might be dependent on the gender of a patient.
In validity, what is regarded an true-experiment?
There is a assignment of random treatments to random patients. It’s all random, so it gives us the most usefull data.
What is conclusion validity?
It reflects how valid the conclusion drawn in retrospect with the significance level of the outcome, etc.
What are threats to conclusion validity?
Unreasonable assumptions or low statisitical power in analysis. Low reliablity of instruments or poor data documentation.
What is internal validity?
How exhaustive and mutually exclusive is the outcome of the results? (are the possible thrid variables that influence the outcome?)
What is contruct validity?
Are the variables that are being analysed correctly reflect the concepts of the study? (f.i. you cant use google searches to indicate strong economical behaviour.)
What is external validity?
How well does our sample group represent a larger population? (generalisation)