Turn variable types, missing values and distributions into a concise description of the cohort.
02
Outcome: a defensible descriptive table and a clear view of the distributions before group comparisons.
Seven-step investigationNow: 02 · Characterise data
Analysis checkpoint
Describe the cohort before comparing groups.
Work through these six prompts in order. Each one supplies what the next one needs.
1 · Before you start
Step 01 must have passed: the participant identifier is unique, variable codes are understood and missing values have been counted.
Words used on this page
Categorical variable
a variable made of groups, such as yes/no or frailty category.
Continuous variable
a measured number that can take many values, such as age.
Distribution
the pattern of values, including centre, spread, skew and unusual observations.
Standard deviation
a measure of spread around the mean, usually shortened to SD.
Interquartile range
the middle 50% of values, usually shortened to IQR.
2 · Why this comes now
Description shows who the findings apply to and whether a mean, median, count or percentage represents each variable honestly. It also reveals problems that should be resolved before group comparisons.
3 · Do this now
Count categories and calculate percentages using the correct non-missing denominator.
For numeric variables, inspect the distribution before choosing mean and SD or median and IQR.
Draft one overall cohort table and a short clinical description of the sample.
Same step in Stata and SPSS
Run Section 02 in Stata or SPSS. Stata uses tabulate and summarize; SPSS uses FREQUENCIES, DESCRIPTIVES and EXAMINE. Match the resulting summaries by variable type rather than by command name.
4 · Read this
Read the valid n and missing count first, then the percentage or measure of centre and spread. Use plots to notice skew, bounds and implausible values.
5 · Ready to continue when
Every important variable has a justified summary, missing denominators are visible and you can describe the cohort in clinical language without a p-value.
Frequency distribution, median/IQR, mean and zeros.
There is no rule that every continuous variable must be reported with a
mean and standard deviation. Match the summary to the distribution and
to what readers need to understand.
Worked description
Build Table 1 before testing it.
The cohort has a mean age of 78.35 years (SD 6.73). Baseline frailty
has a median of 0.213 (IQR 0.154–0.272). Medication count has a median
of 4 (IQR 2–5). At six months, 158 participants are robust, 225
pre-frail and 322 frail; 15 values are missing.
Representative output · same teaching data in all three routes
Overall cohort description
Measure Result
Age, years 78.35 (SD 6.73)
Baseline frailty index 0.213 (IQR 0.154–0.272)
Regular medicines 4 (IQR 2–5)
Six-month frailty category n % of 705 observed
Robust 158 22.4
Pre-frail 225 31.9
Frail 322 45.7
Missing 15 —
What the output means
Age is summarised with a mean and standard deviation because its distribution is sufficiently regular for that description. Frailty and medication count are described with medians and interquartile ranges because they are bounded or skewed. Percentages for six-month frailty use the 705 observed outcomes, while the missing number remains visible.
Write the interpretation like this
Participants had a mean age of 78.35 years (SD 6.73) and a median baseline frailty index of 0.213 (IQR 0.154–0.272). Among 705 participants with observed six-month frailty, 322 (45.7%) were classified as frail; 15 values were missing.
Stata
Use tabulate for categories and summarize, detail for continuous or skewed measures.