The thesis defense is approaching. You have mastered your clinical findings, you know the literature, but are you ready for the committee's statistics questions? These questions come at unexpected moments, and being caught unprepared can negatively affect the defense.
Below are the statistics questions most frequently asked in real defenses and how they should be answered.
Question 1: "How did you determine the sample size?"
Leaving this question unanswered is the most common statistics mistake in a defense.
Ready answer:
"I calculated the sample size by performing a power analysis. Based on [the literature reference], I took the expected effect size as [value]. Accepting an alpha of 0.05 and power of 80%, I calculated with the G*Power program that [N] patients were required. Adding an extra 10-15% for possible loss to follow-up, I reached a total of [N_total] patients."
If you did not use G*Power, calculate it retroactively now and evaluate the power of your findings.
Question 2: "Why did you choose this test?"
Those who cannot justify their test selection appear insecure to the committee.
Ready answer:
"First I performed the normality test. The Shapiro-Wilk test came out p=[value], the data showed a normal distribution [did not show]. For this reason I preferred the parametric [non-parametric] test [test name]."
If you have not checked for normal distribution, do it now and learn the result.
Question 3: "How did you control for confounding factors?"
A question that is always asked, especially in retrospective studies.
Ready answer:
"In my study, [age, sex, comorbidity] were identified as potential confounding factors. Variables with p<0.20 in univariate analysis were entered into a multivariate [logistic/Cox] regression model. The independent effect of each variable was evaluated in this way."
If you did not perform multivariate analysis: "In the limitations section of the study, the inability to control for confounding was stated as a limitation. Propensity score matching may be planned in future studies."
Question 4: "You got P=0.06, how do you interpret this?"
Borderline p values are one of the committee's favorite questions.
Ready answer:
"The p value did not cross the 0.05 threshold, so statistical significance was not achieved. However, a difference that may be clinically meaningful was observed. When a power analysis was performed, it was seen that the study did not have sufficient power to detect this effect size. A study with a larger sample could answer this question more clearly."
Question 5: "Why did you choose [X] as the post-hoc test?"
After the ANOVA is significant, which post-hoc test was chosen is questioned.
Ready answer:
Tukey HSD: "The groups were of equal size and homogeneity of variance was met. Tukey HSD is the most appropriate post-hoc test for these conditions."
Games-Howell: "The Levene test was significant, homogeneity of variance was not met. For this reason I preferred Games-Howell, which does not assume equality of variance."
Bonferroni: "Because the number of comparisons was small, I found the more conservative Bonferroni correction appropriate."
Question 6: "How did you handle missing data?"
There is missing data in every clinical data set. How it was handled is questioned.
Ready answer:
"The missing data rate was determined to be [X]%. [Below 5%]: Because there was a small amount of missing data, complete case analysis was applied, and these data were excluded from the analysis. [Between 5-20%]: Missing data were completed with the multiple imputation method. [Above 20%]: The high rate of missing data is an important limitation of the study and was addressed in the discussion section."
Question 7: "What would you do if your ICC value came out low?"
A question asked in measurement reliability studies.
Ready answer:
"If the ICC value had come out below 0.75, I would have reviewed the measurement protocol. First I would standardize the training of the observers, then I would repeat the measurements. Persistently low agreement indicates that the measurement method itself is not reliable, and in that case the method needs to be changed."
Request a 30-minute free consultation for statistics preparation for your thesis defense.
Where Do People Get Stuck Most in This Analysis?
- The committee will ask "why did you choose a parametric test," but you do not know how to explain the normality test results.
- Your advisor did the power analysis, you only wrote down the result, and you cannot explain the parameters.
- It is unclear how you will present your study's limitations to the committee, and whether you should accept or defend the weak points.