Statistical modelling
Studying this course? Keep your progress in one place.
Key facts
- A model is a simplified description of a real situation, using a random variable and a rule for its probabilities.
- Modelling is a cycle: observe, devise, use, compare, refine and report.
- Typical assumptions: outcomes equally likely, trials independent, constant probability.
- Expected frequency is ; relative frequency is .
- Compare with data: small, patternless differences mean the model is reasonable; large or systematic ones mean it is unsuitable.
What a statistical model is
A model describes a real situation using a random variable and a rule for its probabilities, and works by making simplifying assumptions.
A mathematical model is a simplified description of a real situation that can be analysed using mathematics.
In probability and statistics the model usually describes a situation using a random variable and a rule for its probabilities. For a fair six-sided die, for .
A model cannot capture everything about reality. It works by making assumptions that simplify the situation.
What does a statistical model do?
The modelling cycle
Modelling follows a cycle that is repeated until the model is good enough: observe, devise, use, compare, refine, report.
Modelling follows a cycle, which is repeated until the model is good enough.
You observe a real problem, devise a model with assumptions, use it to make predictions such as expected frequencies, compare the predictions with real data, refine the model if the comparison is poor, and report the findings with the limits of the model.
In an exam, name the step you are performing, such as 'compare the observed frequencies with the model'.
- 1
Observe
a real problem and decide what to find out
- 2
Devise
a model, stating assumptions
- 3
Use
the model to make predictions
- 4
Compare
the predictions with real data
- 5
Refine
the model if the comparison is poor
- 6
Report
the findings, with the limits of the model
Then back to step 1
Predictions from a model disagree strongly with real data. What is the next step in the cycle?
Typical assumptions
Common assumptions are equally likely outcomes, independent trials and a constant probability from trial to trial.
Common modelling assumptions in probability are:
- outcomes are equally likely (a fair die, a fair coin, a random selection)
- trials are independent (one result does not affect the next)
- the probability is constant from trial to trial.
For a spinner with five equal sectors, assumes each sector is equally likely.
For a bus that is late with probability on any day, the model assumes constant probability and independence between days.
Equally likely
- Fair die
- Fair coin
- Random selection
Independent
- One result does not affect the next
Constant probability
- The same probability every trial
Which is a modelling assumption for a bus that is late with probability 0.15?
Expected and relative frequency
If the model gives probability and there are trials, the expected frequency is ; the relative frequency is observed divided by .
If an experiment is carried out times and the model gives probability to an outcome, the expected frequency is .
The relative frequency is .
With more trials, relative frequencies usually settle close to the model's probabilities, so large samples give better evidence.
Compare observed with expected by finding differences, then comment on whether they look small and unpatterned.
- Expected frequency
- Relative frequency
Worked example
A spinner modelled by is spun 200 times. Find the expected frequency of a 3. A 3 occurs 41 times; find its relative frequency.
- 1
Expected frequency .
- 2
Relative frequency .
A fair die is rolled 120 times. What is the expected frequency of a 6?
Evaluating and refining a model
Compare predictions with data, comment in context, and refine by changing probabilities or dropping an assumption.
To judge a model, compare its predictions with data. Always comment in the context of the question.
Refinements include changing probabilities, dropping an assumption such as independence, or collecting more data.
Advantages of models: cheap, quick, allow predictions and 'what if' tests.
Limitations: they simplify, so predictions can be wrong if the assumptions are unrealistic.
Agreement with one data set gives support, not proof.
- 1
Model makes predictions
from its assumptions
- 2
Compare with data
are the differences small and patternless?
- 3
Comment in context
name any assumption that may fail
- 4
Refine the model
change probabilities, drop an assumption, collect more data
Then back to step 1
Small, patternless differences
- Consistent with random variation
- Model is reasonable
Large or systematic differences
- Model is unsuitable
- Name the assumption that may fail
Observed frequencies differ from the model's by large, systematic amounts. What should you say?
Try an exam question
A spinner has five equal sectors numbered 1 to 5. Dan models the score by . (a) State one assumption in Dan's model. (b) The spinner is spun times. Find the expected frequency of a 4. (c) A 4 occurs times. Comment on the suitability of the model.
[4 marks]
- [1]Each sector is equally likely.
- [1]Expected frequency .
- [1]The observed frequency 33 is close to the expected 30.
- [1]So the model appears reasonable, although more data would give stronger evidence.
That's the notes covered.
Carry on to the next subtopic.
Try a question
The model uses for . Which assumption does this model make?
1 mark, marked against the mark scheme
The model uses for . Which assumption does this model make?
1 mark, marked against the mark scheme