Deck 3: The Phoenix of Statistics

Question 1

Your head of clinical practice has followed your advice and now wants you to measure effect sizes. You report a Pearson's r of 0.50 for the impact of increased consumption of oily fish supplements on the symptoms of dementia. Your head of clinical practice wants to know if this is bad, as she remembers that a p-value of 0.30 is not good. What do you tell her?

A) You tell her that effect size and p-values are the same and that a Pearson's r of 0.50 means there is no statistically significant effect and oily fish supplements should not be given to dementia patients.
B) You tell her that effect size and p-values are not the same and that a Pearson's r of 0.50 is a large effect, suggesting she should encourage the use of oily fish supplements in dementia patients.
C) You tell her that effect size and p-values are not the same and that a Pearson's r of 0.50 is a small effect, suggesting she should stop the use of oily fish supplements in dementia treatments until more data analysis is done.
D) You tell her that effect size and p-values are not the same and that a Pearson's r of 0.50 is a medium effect, suggesting she should encourage the use of oily fish supplements in dementia treatments.

Answer 1

You tell her that effect size and p-values are not the same and that a Pearson's r of 0.50 is a large effect, suggesting she should encourage the use of oily fish supplements in dementia patients.

Question 2

In our previous example, the doctor had already calculated the probability of patient life expectancy based on national cancer data. In a Bayesian approach, what sort of probability is this?

A) Positive probability
B) Posterior probability
C) Prior probability
D) Inferior probability

Answer 2

Prior probability

Question 3

A researcher presented a recent study, which showed a statistical significance between increased consumption of sugar and increased tests scores in young children. How can she explain to the chief medical officer that advertising the importance of increasing sugar in children's diets should not be broadcast in the media?

A) A significant result does not mean that the effect is important
B) A significant result means that the effect is strong.
C) A significant result means that the effect is not relevant.
D) A significant result means that the effect is weak.

Answer 3

A significant result does not mean that the effect is important

Question 4

You are the head of a large medical group. You have decided to adopt a Bayesian approach to data analysis and modelling. When you announce this new policy, your staff are unhappy and unconvinced, as they are used to a NHST approach. You stress that the Bayesian approach has several key advantages, including which of the following.

A) You can evaluate statistical significance using p-values.
B) You can evaluate the likelihood of the null hypothesis being true.
C) You can reject null hypotheses without any data collection.
D) You can evaluate complex statistical models without data.

Question 5

You lead a product-testing unit for a large pharmaceutical company. Your team has conducted forty trials of a new antibiotic but you are not sure if the results are conclusive enough to urge the company to start producing the new drug. A new data analyst has joined your team suggesting that meta-analysis might be a good idea, do you agree?

A) Yes, because the forty trials were identical and tested the same research question and therefore we can calculate an average significance for the new drug.
B) No, because the forty trials were identical and tested the same research question we cannot calculate an average effect size for the new drug.
C) Yes, because the forty trials were identical and tested the same research question and therefore we can calculate an average effect size for the new drug.
D) No, because the forty trials were identical and tested the same research question we cannot calculate an average significance for the new drug.

Question 6

A midwife conducted a Bayesian analysis of stork numbers and declining birth rates. She calculated a Bayes factor of 1. Should she use stork numbers to predict birth rates?

A) No, a Bayes factor of 1 suggests that although declining stork numbers appear to coincide with declining birth rates it is not 'worth mentioning' and other factors may be at work.
B) Yes, a Bayes factor of 1 suggests that it is worth trying to increase stork numbers in order to increase the birth rate.
C) No, a Bayes factor of 1 suggests that the data is corrupted.
D) Yes, a Bayes factor of 1 suggests that it is worth heavily investing in reducing stork numbers in order to reduce the birth rate.

Question 7

A researcher was assessing patient healing time when using a new form of surgical dressing called MediDressX. He had a sample size of 48 and a p-value of 0.20. Does the researcher recommend that hospitals stop using this product?

A) No, because the sample size is large and therefore the p-values are accurate.
B) Yes, because statistical significance has nothing to do with sample size.
C) No, because the sample size is small and p-values are easily affected by sample size.
D) Yes, because the sample has low confidence levels.

Question 8

A trainee data analyst for a large hospital, which has higher than average patient mortality rates, has just completed an examination of key factors that may be influencing these high mortality rates. However, he finds only one statistically significant factor, which he includes in his report but deliberately, omits the other six non-significant findings. What is the term for what the data analyst has done?

A) HARKing
B) Meta-analysis
C) p-hacking
D) Bayesian analysis

Question 9

A researcher was interested in assessing gender and response to pain, so conducted a t-test. The mean for males was 66.25 and the mean for females was 78.24, with both groups having a standard deviation of 7. What is the effect size using Cohen's d?

A) -1.712
B) 1.7
C) 0
D) 1.712

Question 10

You have joined the data modelling team for a health policy unit. Your boss has decided that from now on the team will adopt a Bayesian approach. However, not all staff understand what this is; your boss asks you to present a training session. How would you explain s Bayesian appaorch in your session introduction?

A) An approach that allows you to update the likelihood of your statistical model as more data is collected.
B) An approach that allows you to focus on testing the null hypothesis based on data collection.
C) An approach where you do not modify the likelihood of your statistical model as more data is collected.
D) An approach where you reject your statistical model once data is collected.

Question 11

Which of the following is not a factor in researchers' over-use of p-values and NHST in health research?

A) Time constraints within research encourages quick results.
B) Pressure to get a significant result that healthcare managers can easily understand and apply.
C) Career bonus structures incentivise 'results'.
D) Statistical software encourages the over-use of p-values.

Question 12

A doctor is concerned about the life expectancy rates of cancer patients in his clinic. There were forty patients in the clinic all of whom were in remission from cancer. Theoretically, all the patients had an equal probability of living for at least ten years post-treatment. However, the doctor had data that suggested that younger patients were more likely to live longer than older patients. This data was from national cancer data. However, the doctor had used the same and newly developed approach to treatment for all his patients and therefore wanted to determine the probability that all patients would live for at least ten years post-treatment, irrespective of their age. What formula could he use to determine this probability?

A) NHST
B) Pearson's r
C) Cronbach's Alpha
D) Bayes' theorem.

Question 13

Your health research lecturer has devoted the past ten weeks to teaching you the Bayesian approach and is now asking that you offer a critique of it. What key criticism could you raise?

A) The lack of reliance on a prior probability is overly subjective and therefore can be open to a researcher's degrees of freedom.
B) The lack of reliance on a null hypothesis is overly objective and therefore open to a researcher's degrees of freedom.
C) The reliance on a prior probability is overly objective and therefore not open to a researcher's degrees of freedom.
D) The reliance on a prior probability is overly subjective and therefore can be open to a researcher's degrees of freedom.

Question 14

Your head of clinical practice has just read a book on criticisms of the NHST and worries that all clinical data analysis is now flawed. How might you reassure her?

A) NHST does have its flaws but if we incorporate an examination of effect sizes into our analysis, we should be able to trust our research outputs.
B) NHST does have its flaws but everyone else uses it, therefore we should.
C) NHST is a flawless approach and the book was probably written by a disciple of the Bayesian approach.
D) NHST is a flawless approach and we need to invest in more data analysts who are trained in it.

Question 15

A person involved in health promotion wanted to examine the likelihood that health promotion messages on social media would be more effective than in print media. She conducted one study where the probability of making a Type I error was 0.05 and a Type II error was 0.2. Does her research have empirical probability?

A) No, to have empirical probability the likelihood of an effect being detected requires a series of repeated identical experiments, where the probability of making a Type I error is above 0.05 and a Type II error is 0.2.
B) No, to have empirical probability the likelihood of an effect being detected requires a series of repeated identical experiments, where the probability of making a Type I error is 0.05 and a Type II error is 0.2.
C) Yes, to have empirical probability the likelihood of an effect being detected requires a single experiment, where the probability of making a Type I error is above 0.05 and a Type II error is 0.2.
D) No, to have empirical probability the likelihood of an effect being detected requires a single experiment, where the probability of making a Type I error is above 0.05 and a Type II error is 0.1.

Question 16

A researcher was interested in patient use of pop-up Flu vaccine clinics across thirty sites nationally. Different researchers collected and analysed data across each of the sites but the resultant thirty reports showed differing p-values, some sites found a statistical significance between opening hours of clinics and patient usage and others did not. Which of the following would is useful for her to review?

A) Confidence intervals
B) Levels of missing data
C) Outliers
D) The Null Hypotheses