Direct answer
A worked example of Data Surprise shows, step by step, how you turn an expectation about data into a single quantified “surprise” number after you observe the actual result. The example must state every assumption that affects the calculation, because the size and meaning of “surprise” depends on the baseline model and inputs you choose.
Mechanism or definition
Data Surprise is a general idea: you have (1) an expectation about some data outcome and (2) you observe a different result. The “surprise” is the difference between expected behavior and realized behavior.
To make this concrete, you must define a method that converts “unexpectedness” into a number. Common ways include:
- Difference from a baseline: surprise = observed − expected.
- Standardized surprise: surprise = (observed − expected) / expected variability, so surprise is comparable across contexts.
- Tail-probability surprise: surprise = how unlikely the observed value is under the expectation.
In a worked example, you also have to decide what information counts as “expected.” For instance, “expected” might mean a forecast, a consensus estimate, or a model prediction made before the data is released. The key is that the example must specify the expectation source in a hypothetical, self-contained way.
Evidence or example
Here is one fully stated numerical worked scenario that illustrates the mechanics, without assuming real-time markets.
Step 1: Set the “expected” and its variability
Assume a dataset reports a quarterly growth rate. Before the release, you model the growth rate as follows:
- Expected value (mean): 2.0%
- Expected variability (standard deviation): 0.5%
Assumption A: these numbers represent what you believe will typically happen if your expectation is correct.
Step 2: Observe the “real” value
Suppose the released growth rate is:
- Observed value: 3.0%
Assumption B: “observed” is the data value you measure after release.
Step 3: Compute a standardized Data Surprise
Compute the standardized surprise:
- Surprise = (observed − expected) / standard deviation
- Surprise = (3.0% − 2.0%) / 0.5%
- Surprise = 1.0 / 0.5 = 2.0
Interpretation (under this model): a standardized value of 2.0 indicates the result is “two variability units” away from your expectation.
Step 4: Translate surprise into a qualitative meaning (optional)
To attach a notion of rarity, you must also assume a distribution shape (for example, roughly symmetric and bell-shaped under the expectation). If you make that assumption, a standardized surprise of 2.0 corresponds to a tail event. The exact rarity depends entirely on the chosen distribution.
Assumption C: you use a specific distribution to map standardized surprise to “how unlikely.”
What this illustrates
- The same observed value could be “more” or “less” surprising depending on the expected mean and variability.
- The method turns “unexpectedness” into a number that you can compare across different releases, but only if the assumptions are consistent.
Limitations and risks
- Model and assumption dependence: If the expected variability is too small, surprises look larger than warranted; if too large, surprises look smaller.
- Distribution mismatch: If real data is skewed, heavy-tailed, or has regime changes, tail-probability interpretations can be misleading.
- Context and costs: Even if the data surprise is computed correctly, real-world outcomes may not follow the computed direction because transaction costs, execution timing, and other information can matter.
- Failure mode: confusing explanation with prediction: Historical relationships between “surprise” and later outcomes do not automatically establish that future surprises will cause similar effects.
Verification or next question
To independently verify a “Data Surprise” worked example, you can:
- Recompute the arithmetic from the stated expected value, variability, and observed outcome.
- Check that the definition of surprise matches the formula you used (difference vs standardized vs tail probability).
- Verify that the distribution assumption (if any) is explicitly stated, because it changes any “rarity” interpretation.
If you want to compare two worked examples, the next question is: Do they use the same expectation method, variability definition, and distribution assumptions? If not, their surprise numbers may not be comparable.