Hersi Maths WhatsApp me

Understand · explore · practise

Trace, missing data and weather samples

Distinguish trace rainfall, measured zero and missing data. Compare cleaning decisions with an original synthetic table and explain their effects on means and sample size.

Before you startWeather data context, means and sampling frames.

01 / Three records, three meanings

Trace is small; missing is unknown.

0: a recorded zero. tr: trace rainfall below 0.05 mm. n/a: unavailable.

Do not turn missing information into a measured zero.

A trace entry records a tiny amount, not an absent observation. It usually needs a stated numerical approximation for arithmetic. A missing entry provides no rainfall amount for that day.

Pearson’s 2023 examiner report, question 3 explains why discarding trace days changes calculations. Follow the convention specified by the question; otherwise state and justify your approximation.

Keep track of the denominatorExplore

Invented teaching data, not Met Office records. Daily rainfall (mm):

DayRecord
10
2tr
30.6
4n/a
51.4
6tr
70.2
80

Total 2.2 mm ÷ 7 observed days ≈ 0.314286 mm/day.

Trace is approximated as zero; both trace days remain in the count. The missing day is excluded.

Numerical displays are rounded to six decimal places. A mean over observed days is not a known mean for all eight days.

02 / Replace trace transparently

Keep the day even when you approximate its rainfall.

Use the invented eight-day table with trace approximated as zero.Worked example

Known numerical total: 0 + 0.6 + 1.4 + 0.2 + 0 = 2.2 mm

The two trace entries contribute zero under this approximation.

There are seven observed days

Count the two trace days; exclude only the n/a day.

Mean ≈ 2.2/7 = 0.314286 mm/day

This is the observed-day mean under the stated convention.

Using 0.025 mm for each trace is a midpoint-style approximation to a value below 0.05; it is not a claim to know the actual amounts. Here it gives 2.25/7 ≈ 0.321429 mm/day. State which convention you use.

Watch: trace days still count

Pause, replay or seek freely. The notes explain the same idea and stay in view.

01 · Count usable days

A sample has 12 dates: eight numerical rainfall values, three traces and one unavailable entry. With trace treated as zero, how many days enter the mean?

Hint

Trace days have an observation; the unavailable day does not.

Worked solution

Eleven: eight numerical records plus three trace records.

02 · Calculate with a convention

The eight numerical values total 5.5 mm. Find the eleven-day mean if each of the three traces is approximated as zero.

Hint

Divide by the observed-day count, not the numerical-cell count.

Worked solution

5.5/11 = 0.5 mm/day.

03 / Missing is not zero

An unavailable value could be small or large.

If you calculate a mean from the observed entries, report the number used and explain which entries were unavailable. Replacing n/a with zero silently assumes no rain fell, which the record does not say.

Missingness may be related to the measurement process or weather. Excluding missing days can therefore affect representativeness; it is a transparent calculation rule, not a guarantee of an unbiased estimate of all days.

03 · A false zero

Observed rainfall values on four days are 0, 0.4, 0.8 and 2.0 mm; a fifth is unavailable. Find the observed-day mean and explain why dividing by five is unjustified.

Hint

The known total is 3.2 mm.

Worked solution

3.2/4 = 0.8 mm/day. Dividing by five gives 0.64, implicitly treating the unknown rainfall as zero.

04 · Unknown full-period mean

Can the exact five-day mean be recovered from those four values and n/a?

Hint

The missing amount contributes to the total.

Worked solution

No. If the missing amount is x mm, the five-day mean is (3.2 + x)/5. Its value is unknown without more information.

04 / How much can traces change a mean?

A small approximation can be bounded.

With r traces and n observed days, replacing every trace by zero changes the mean by less than 0.05r/n mm.

This bound concerns trace approximation, not unknown missing values.

In the model, r = 2 and n = 7, so the change is less than 0.1/7 ≈ 0.014286 mm/day. This explains why a declared zero approximation can be reasonable for some purposes. It does not justify discarding those two observed days.

05 · Bound the change

There are four trace entries among 20 observed days. Bound the difference between the exact observed-day mean and the mean obtained by replacing all traces with zero.

Hint

Each omitted trace amount is below 0.05 mm.

Worked solution

The total difference is less than 0.2 mm, so the mean difference is less than 0.2/20 = 0.01 mm/day.

06 · Why discarding differs

In the model, discarding both trace days gives 2.2/5 = 0.44. Explain why this is not the same as approximating trace as zero.

Hint

The denominator changed.

Worked solution

Approximating retains seven observed days, giving 2.2/7. Discarding removes two low-rainfall observations and changes the population of days being averaged.

05 / Requested versus usable samples

State the frame and missing-data policy before selection.

Requesting 30 randomly selected calendar dates does not guarantee 30 usable readings for a variable. You could sample from all eligible dates and report missingness, or deliberately define a frame of dates with that variable available. The latter changes the population directly represented.

If comparing two variables, use records where both needed values are available and retain the pairing. Different variables can have different usable sample sizes.

07 · The actual sample size

Twenty-five dates are chosen. Temperature is available for 24 and rainfall for 23, but two different dates are missing rainfall and a third is missing temperature. How many complete temperature–rainfall pairs remain?

Hint

Exclude the union of missing dates.

Worked solution

Three distinct dates lack at least one variable, leaving 25 − 3 = 22 complete pairs.

08 · Redefining the frame

A student removes all missing-wind-speed dates before selecting a random sample. What population is directly sampled?

Hint

The eligible frame now contains only available readings.

Worked solution

The dates with wind-speed readings available in the chosen station and period. The sample is not automatically representative of dates with unavailable readings.

06 / Cleaning is a documented decision

Check definitions before changing records.

Preserve the original table. Create a working copy or an additional cleaned column, record trace substitutions and identify missing values explicitly. Inspect unusual values against units and context. An extreme but valid observation should not be removed merely because it changes the mean.

09 · A large rainfall entry

A rainfall value is much larger than nearby entries. Should it automatically be deleted?

Hint

Unusual is not the same as erroneous.

Worked solution

No. Check the source, units and plausibility. Retain a valid extreme value; if an error is established, document the correction or exclusion.

07 / State value rules and counts

A reproducible mean needs more than a calculator answer.

  1. Identify zero, trace and unavailable entries.
  2. Follow the question’s convention, or state a justified trace approximation.
  3. Keep trace days in the observed-day count.
  4. Do not silently replace missing values with zero.
  5. Report the usable sample size and preserve paired records.
  6. Limit conclusions to the chosen frame and period.

10 · Report the model result

Write a defensible one-sentence report for the model using trace as zero.

Hint

Include the approximation, count and missingness.

Worked solution

For the seven observed days in this invented eight-day table, the mean rainfall is approximately 0.314 mm/day when both traces are treated as zero; one unavailable day is excluded, so this is not the known mean for all eight days.

Section 1 of 7 · Three records, three meanings