P3-7.1 What Should We Compare the Structure We Kept Against So That Change Becomes Visible¶

Section ID: P3-7.1 Version: v2026.09.20

A current value alone cannot tell us whether it rose or fell relative to the past. Here a baseline is the reference value paired with it. This section connects recent and baseline tables to create difference columns and review statements. The distinction from a baseline model for performance evaluation is discussed in P3-7.3.

Align Conditions Before Joining Recent and Historical Tables¶

This is a fictional aggregation example. The recent period is 2026-09-20 from 09:00 inclusive to 09:30 exclusive; the reference period covers the same daily half-hour from September 13 through 19. Include actions completed within each interval. Assume equipment and operating conditions other than process type match, as do sensor location, units, middle-segment definition, and missing-data rules.

First calculate each action's middle-segment mean flow, then average actions equally within each group. Units are L/min. This is not a pooled average of all raw measurement points, and the numbers below cannot reconstruct raw logs or their spread.

The baseline uses records collected before the recent period starts. This period is chosen to answer a question about earlier actions at the same time of day. Older records or changed operating conditions require checking whether the reference still answers that question.

Joining by Process Type Creates a Difference Column¶

Join the tables by process_type and calculate diff = recent_mid_flow − baseline_mid_flow. Each type has one baseline row here. If type alone does not identify matching conditions in real data, include equipment, operating mode, measurement definitions, or other necessary join conditions.

For A, 2.10−2.45=−0.35 L/min, giving “The recent actions' middle-segment mean flow is 0.35 L/min below the selected historical reference.” For B, 2.30−2.28=+0.02 L/min. This table alone cannot establish whether a difference is acceptable, random variation, or a failure. A small difference and a normal state are different judgments.

If a new type has no reference, do not replace its baseline with zero and subtract. Mark the comparison unavailable and obtain a suitable reference. If several baseline rows match, choose the intended period and conditions rather than accidentally multiplying a recent row through the join.

flowchart TD
    A[Recent window] --> C[Comparison table]
    B[Baseline window] --> C
    C --> D[Difference columns]
    D --> E[Human review sentence]

Changing the Baseline Changes the Comparison Question¶

Reversing A's subtraction gives 2.45−2.10=+0.35. That is a column with the opposite definition, not an arithmetic error. Record “recent−baseline” alongside the name diff.

Comparing the same recent value 2.10 with a reference of 2.00 L/min from another historical period gives +0.10 L/min. Here 2.00 is hypothetical, not calculated from the reference period above. Actual use would require separately reporting its period, conditions, aggregation rule, and count.

Choosing a baseline to obtain a desired sign changes the question about differences from a selected past period while presenting it as the same conclusion. If both references are used, identify their respective populations and report the differences separately. A baseline is not a number selected after seeing which result looks preferable.

Check the arithmetic: adding baseline 2.45 to A's −0.35 recovers the recent value 2.10. For B, 0.02+2.28=2.30. Values, signs, and units must agree for someone to reproduce the review statement.

The 2.8 L/min baseline in P3-7.2 belongs to a separate fictional example using late-segment means and 200 historical actions. Do not combine it with this middle-segment table as if it were the same dataset.

Checklist¶

• Can you explain the periods, inclusion conditions, units, and aggregation units of both tables?
• Did you join recent and reference rows with matching conditions?
• Can you check the sign and L/min units of recent−baseline?
• Can you explain how changing the reference changes the population and question?

Sources and Further Reading¶

• U.S. Bureau of Labor Statistics, Base period. Because it explains a base period as a reference for comparing other times, it supports the central point of this section that change becomes visible only when the current range and the baseline range are placed together. https://www.bls.gov/bls/glossary.htm / Accessed: 2026-07-20
• National Cancer Institute, baseline. Because it explains baseline as the standard against which later change is compared after an initial measurement is set, it provides a general basis for reading baseline in Part 3 as a state-comparison reference. https://www.cancer.gov/publications/dictionaries/cancer-terms/def/baseline / Accessed: 2026-07-20