RS517 Chap.4 Descriptive Statistics and Missing Data
Descriptive Statistics and Missing Data
Describe before testing
The mean uses every value and is pulled by tails, while the median identifies the ordered middle and resists extremes. Standard deviation describes spread around a meaningful mean; interquartile range describes the middle half of ordered values.
Variable type and distribution determine the honest combination.
Plots reveal model problems
Histograms, boxplots and scatterplots show skew, clusters, floor effects, unequal spread and influential observations that a table can conceal. An unusual value should be checked against the source and interpreted before any exclusion decision.
Missingness has a mechanism
The amount missing is only a start.
Missingness may be unrelated to study data, explained by observed variables or tied to unseen values. Observed-data methods can support the second situation but cannot verify it, while sensitivity analysis is needed when unseen outcomes may drive loss.
Let the distribution choose the summary
Begin with variable type, units and a plot.
Counts and proportions describe categories; ordered categories require care because adjacent codes may not represent equal distances; continuous variables permit arithmetic summaries but can still be skewed, bounded or multimodal. The mean answers a balancing-point question and uses every observation, whereas the median answers an ordered-middle question and resists long tails.
Pair standard deviation and interquartile range with compatible centres, and retain the sample size and denominator.
Investigate unusual values without shopping for results
An extreme observation may be an entry error, a unit error, a genuine population member or a sign that the proposed model is unsuitable. Verify the source record, plot the value in context and examine influence under the prespecified analysis.
Sensitivity analysis can show whether a conclusion depends on the observation, but deleting it solely because a significance threshold changes is indefensible. If a transformation is used, interpret the fitted model on its transformed scale and supply an original-scale summary for practical meaning.
Make the missingness argument explicit
The percentage missing does not identify the mechanism.
Ask when the value became unavailable, which recorded variables predict that event and whether the unseen value itself could still drive loss. Complete-case analysis changes the represented sample unless its assumptions hold. Multiple imputation or likelihood methods can use observed predictors and preserve uncertainty, but neither proves that unseen outcomes are irrelevant.
Sensitivity analysis should vary the untestable part of the assumption and show how the conclusion moves.
Revision drill
For a small dataset, draft a data dictionary, choose a plot and justify a centre-spread pair. Create a cleaning log for range, logic, duplicate and derivation checks.
Then invent two different reasons for the same missing pattern and explain why they demand different conclusions even though the percentage unavailable is identical.
What this chapter covers
- 01
Variable scales and suitable summaries
- 02
Mean, median and distribution shape
- 03
Standard deviation and interquartile range
- 04
Outliers and influence
- 05
Missing-data patterns and mechanisms
- 06
Complete-case, model-based and sensitivity approaches
Worked example · free
Summarise a skewed completion time
- 3Identify skew and the typical-value question.
- 5Choose median and interquartile range, preserving the extreme case.
Key terms
- Median
- The middle ordered value, or average of the two middle values when the count is even.
- Interquartile Range
- The distance between the first and third quartiles, containing the middle half of observations.
- Standard Deviation
- A measure of spread around the mean on the original variable scale.
- Influential Observation
- A value whose removal or movement changes a fitted result substantially.
- Sensitivity Analysis
- An analysis that tests how conclusions change under alternative assumptions or data decisions.
Descriptive Statistics and Missing Data FAQ
When should the median replace the mean?
Use the median when the typical ordered value is the target and a long tail or extreme values make the mean unrepresentative. Keep the distribution visible and report an interquartile range. The mean may still matter for totals, costs or models, so the decision follows the question.
Should an outlier be deleted?
Not because it weakens significance or normality. Verify units, data entry and the source record, assess whether the value is possible, inspect its influence and run a justified sensitivity analysis. If the conclusion changes, report both results and explain the population represented by each.
Why is complete-case analysis risky?
It analyses only records with all required values. When completeness is related to prognosis, group or outcome predictors, that subset can distort the estimate and waste precision. Describe missingness, include relevant observed predictors in an appropriate method and test sensitivity to unseen outcomes.
Exam move
For every dataset, write variable type, plot, centre and spread before running an inferential test. Create a missingness table by variable, group and baseline predictor. Practise explaining one result twice: once under an observed-data assumption and once under a plausible adverse sensitivity scenario.
Working through Descriptive Statistics and Missing Data in RS517? Sia is AskSia’s AI Statistics tutor — ask any RS517 Descriptive Statistics and Missing Data question and get a clear, step-by-step explanation grounded in how RS517 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.