A comparable baseline
9 min read
Lesson objectives
- Apply A comparable baseline to a fictional community case, documenting the decision and its limits in a reviewable work product.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
After defining a bicycle-loss problem, a facility manager may hand you two seemingly conflicting figures: 12 events in the facility log for a month and nine in an official reporting system. Do not add them to make 21 or select the larger figure to make the case for action look stronger. Some people may have reported to both systems, and the systems may apply different recording rules. Before plotting a trend, write a measurement dictionary: eligible behavior, counting unit, event date rather than message date where available, place boundary, source, duplicate rule, and treatment of later corrections. That dictionary is part of the baseline, not a technical afterthought.
Take fictional data from the facility system alone. In the first four weeks it records 12 independent losses and 600 bicycle parking episodes at the east entrance. In the next four weeks it records 15 losses and 1,000 parking episodes. The raw count rises by three, which is 25% of the initial 12. The descriptive measure “loss events per 100 parking episodes” is 12/600×100=2.0 for the first period and 15/1,000×100=1.5 for the second. It falls by 0.5 event per 100 episodes, or 25% relative to its first value. Do not claim that a particular person's probability of bicycle theft fell by 25%. An episode is not a unique person; some people park repeatedly, some losses may go unreported, and no design here establishes why the measure changed.
A useful denominator represents opportunities for the numerator event to occur. If the numerator is loss of bicycles parked at the east entrance, parking episodes there in the same period offer a practical initial denominator. The city's population does not tell us how many bicycles faced conditions at that entrance. If management moved parking to another entrance in the second period, separate the sites. Combining them could make a tidy ratio that compares different environments. Check that the episode counts cover the same opening hours as the loss records and that a new sensor did not change the counting method. When exposure cannot be estimated, present a count with its limits rather than calling it a risk rate.
Prepare a small table before making claims: period, place boundary, bicycle-loss inclusion rule, independent events, parking episodes, division result, source of numerator and denominator, and any change in reporting. If a late message describes an event from period one, assign it to the event date and record when the table was amended. If the event date cannot be established, mark it unknown rather than silently using the message date. Retain an internal event identifier for deduplication, while excluding the bicycle owner's name from a public or learning report.
The official series is useful but addresses a different measurement question. Its report threshold or classification may differ from the facility log, and willingness to report can change. UNODC teaching material explains that official records omit some victim experiences and that reporting varies. Label each source's series and examine overlap before using them as complementary evidence; they are not automatically additive. When the measure changes later, ask whether exposure changed, reporting changed, events changed, or several of these occurred together. A preferred story is not evidence.
Decide whether the numerator counts events or affected bicycles. One occurrence may involve two bicycles belonging to different people: that is one event but two lost bicycles. The choice depends on the decision. Event counts may help a place manager examine recurring conditions; bicycle counts convey a different measure of property loss. Do not change units halfway through a series. With small counts, moving just one event can noticeably change a ratio, so display the original counts beside the calculated measure. The move from 2.0 to 1.5 in this example is a description, not a statistical test or proof that an improvement will persist.
Finally, check arithmetic and meaning separately. Recalculate from the original counts. Ask a colleague to interpret the denominator and unit from the column heading alone. If they read “per 100 people” when you mean “per 100 parking episodes,” rewrite the label. Do not compare a four-week count with a one-week count without aligning or explaining the periods; a longer observation window can have more events while the daily pattern stays constant. The specialist's deliverable is a reproducible baseline table with inclusion decisions and a statement of what the figures cannot prove.
Test the denominator when part of the observation schedule is missing. An employee counted parking operations in nine of twelve planned observation periods. If the loss reports cover all twelve periods, combining them with the denominator from nine overstates the measure. Select a matching window for events and exposure, or show the count and explain that exposure coverage is incomplete. This preserves the meaning of the fraction before comparison begins.
Draw a simple calendar with days as rows and morning and evening periods as columns. Mark observed cells and explain missing cells in plain language. Missing observations may coincide with severe crowding that prevented counting. In that case the missing portion differs from the observed portion, and using the observed average requires particular caution.
Improve future collection operationally. Separate exposure counting from receiving reports during busy hours, or preselect shorter periods across the schedule when resources are limited. Label the result as an estimate from a time sample whenever sampling is used. Assigning a replacement observer is preferable to leaving a blank that someone later reads as zero.
Comparing baselines also requires checking the definition of a parking operation. Does it mean every arrival at a rack or every bicycle visible when the observer looks? One bicycle arriving twice represents two uses, whereas a snapshot measures occupancy at a particular moment. Choose the definition that answers the prevention question and state it in the collection form. Changing it between periods breaks comparability even if the table layout remains identical.
Reconstruct a small calculation from the original record. Link each numerator, time window and denominator to a named source, then ask a colleague to reproduce the result without oral explanation. If the colleague must guess what “attendance” means, improve the dictionary. This checks measurement before discussing intervention effects; it is not a request to gather more information without purpose. Present the result in an understandable unit, such as documented events per hundred observed parking operations during specified hours.
Lesson summary
An official list contains nine events for the first period.