01. Read results with denominators, distributions and uncertainty
Lesson
LEARNING OBJECTIVES
Extract the population and analysis denominator, judge whether summaries fit the variables, and distinguish precision, effect size and freedom from bias.
READ THE TABLE BEFORE THE CONCLUSION
Identify who entered the study, who contributed each outcome and why observations are missing. A sample of 500 participants does not mean every result has denominator 500. Different follow-up periods or exclusions may make seemingly similar percentages answer different questions.
For categorical variables, extract counts, denominators and category definitions. For quantitative measurements, check units, distribution and the rationale for mean/SD or median and other summaries. A numeric ID or category code does not become a meaningful continuous measure because software can average it.
WORKED FICTIONAL DISTRIBUTION
Five invented waiting times are 5, 6, 7, 8 and 54 minutes. The mean is 16 minutes and median 7 minutes. Reporting only the mean conceals the shape; reporting only the median can hide a very long wait. Ask whether the extreme value is credible and show the spread appropriately rather than deleting it automatically.
Now imagine 40 participants out of 80 answering an outcome item reported a service barrier, while 20 of the original 100 did not answer. The observed proportion among respondents is 50%, not evidence that exactly half of all eligible people experience the barrier. Report missingness and consider its possible influence.
INTERPRET ESTIMATES AND INTERVALS TOGETHER
Extract the actual measure—risk difference, RR, OR, mean difference or another quantity—and its scale, reference group, period and uncertainty. An OR is not automatically an RR. A narrow confidence interval describes precision under assumptions, not protection from selection or measurement bias.
A confidence interval including a null value does not prove no meaningful effect. Examine which effects remain compatible with the data and model. Statistical significance does not measure practical importance, and a p-value is not the probability that the null hypothesis is true.
YOUR TASK
Choose an accessible quantitative paper. Build an extraction card containing design, population, primary outcome, exposure/comparison, analysis n, missingness, units, estimate, interval, adjustment and stated limitations. Locate a sentence where the conclusion could be stronger than the reported result supports.
Rewrite that sentence to name the effect, uncertainty and main limitation without substituting “significant” for a scientific interpretation.
SELF-CHECK AND FEEDBACK
1. Is median 7 proof that most waits were acceptable? No; acceptability requires context and the whole distribution matters.
2. Does missingness always make the result biased? Not automatically, but its process and analytical handling require assessment.
3. Does a small p-value establish importance or causation? No; inspect the design, magnitude and assumptions.
PORTFOLIO OUTPUT
A results-extraction card and a revised interpretation paragraph. Use the core epidemiology course for calculation practice; this lesson emphasizes appraisal.
FURTHER READING
ASA Statement on p-Values.
https://doi.org/10.1080/00031305.2016.1154108
CDC: Analyzing and Interpreting Data.
https://www.cdc.gov/field-epi-manual/php/chapters/analyze-interpret-data.html
Extract the population and analysis denominator, judge whether summaries fit the variables, and distinguish precision, effect size and freedom from bias.
READ THE TABLE BEFORE THE CONCLUSION
Identify who entered the study, who contributed each outcome and why observations are missing. A sample of 500 participants does not mean every result has denominator 500. Different follow-up periods or exclusions may make seemingly similar percentages answer different questions.
For categorical variables, extract counts, denominators and category definitions. For quantitative measurements, check units, distribution and the rationale for mean/SD or median and other summaries. A numeric ID or category code does not become a meaningful continuous measure because software can average it.
WORKED FICTIONAL DISTRIBUTION
Five invented waiting times are 5, 6, 7, 8 and 54 minutes. The mean is 16 minutes and median 7 minutes. Reporting only the mean conceals the shape; reporting only the median can hide a very long wait. Ask whether the extreme value is credible and show the spread appropriately rather than deleting it automatically.
Now imagine 40 participants out of 80 answering an outcome item reported a service barrier, while 20 of the original 100 did not answer. The observed proportion among respondents is 50%, not evidence that exactly half of all eligible people experience the barrier. Report missingness and consider its possible influence.
INTERPRET ESTIMATES AND INTERVALS TOGETHER
Extract the actual measure—risk difference, RR, OR, mean difference or another quantity—and its scale, reference group, period and uncertainty. An OR is not automatically an RR. A narrow confidence interval describes precision under assumptions, not protection from selection or measurement bias.
A confidence interval including a null value does not prove no meaningful effect. Examine which effects remain compatible with the data and model. Statistical significance does not measure practical importance, and a p-value is not the probability that the null hypothesis is true.
YOUR TASK
Choose an accessible quantitative paper. Build an extraction card containing design, population, primary outcome, exposure/comparison, analysis n, missingness, units, estimate, interval, adjustment and stated limitations. Locate a sentence where the conclusion could be stronger than the reported result supports.
Rewrite that sentence to name the effect, uncertainty and main limitation without substituting “significant” for a scientific interpretation.
SELF-CHECK AND FEEDBACK
1. Is median 7 proof that most waits were acceptable? No; acceptability requires context and the whole distribution matters.
2. Does missingness always make the result biased? Not automatically, but its process and analytical handling require assessment.
3. Does a small p-value establish importance or causation? No; inspect the design, magnitude and assumptions.
PORTFOLIO OUTPUT
A results-extraction card and a revised interpretation paragraph. Use the core epidemiology course for calculation practice; this lesson emphasizes appraisal.
FURTHER READING
ASA Statement on p-Values.
https://doi.org/10.1080/00031305.2016.1154108
CDC: Analyzing and Interpreting Data.
https://www.cdc.gov/field-epi-manual/php/chapters/analyze-interpret-data.html
My study notes
Stored only in this browser. Do not enter patient information.Saved in this browser. No account needed.