Sale label saturation statistics and demand shifted between products

Updated 13 source citationsSources checked 13 November 2026

On this page
  1. Quick answer
  2. Key Takeaways
  3. What changed when more products carried labels?
  4. How should the demand indices be read?
  5. Does the test prove that sale labels lost credibility?
  6. What should a store measure before adding more badges?
  7. What baseline mix would make the reported changes balance?
  8. Why might the contribution result differ from the unit result?
  9. Which calculations can a reader reproduce?
  10. Methodology and Sources
Download study text

Quick answer

Sale label saturation statistics show why a promotional badge can help one product while weakening another. In a catalog field test, adding sale claims to five products raised their indexed demand from 100 to 157. Demand for three other products that already had sale claims fell from 100 to 38. Prices were unchanged between versions. (Anderson and Simester, May 2001)

The decision is whether to label more of an assortment as on sale. The local product gains are real study outcomes, but their row-specific indices cannot be added into a category growth calculation. This analysis retains that restriction and distinguishes a badge gaining attention for an item from a promotion increasing demand for the whole selection.

Key Takeaways

  • Newly labeled products sold 57% more units at unchanged prices. (Anderson and Simester, May 2001).
  • Products already carrying sale labels sold 62% fewer units when more nearby items were labeled. (Anderson and Simester, May 2001).
  • The experiment did not establish a significant aggregate demand gain across the eight products. (Anderson and Simester, May 2001).

What changed when more products carried labels?

Two catalog versions went to samples of 67,000 US households each. The authors deliberately conceal the actual product category; this article therefore refers to the items as products. On the focal pages, three of eight products carried a sale cue in control and all eight in treatment. The five added claims were the intervention. The three existing claims and the prices of all eight products stayed the same. One accessory had no sale label in either condition. (Anderson and Simester, May 2001)

Sale label saturation statistics: newly labeled product demand had an index of 157, previously labeled products 38 and the unlabeled accessory 90 when all eight products carried labels; each control row equals 100.
Figure 1: Newly labeled items gained demand while already labeled items lost it, and the normalized rows cannot establish total category growth. Source: Anderson and Simester, May 2001. Original comparison from the cited study. Study period and calculation limits appear in the text.

The catalog used a preseason sale description. This was not a test of a new discount amount. A current merchant should not assume that a label is equivalent to reducing the price or disclosing a genuine reference price. The shopper may be reacting to the signal, the assortment context or the apparent scarcity of the promotion.

Assignment used zip codes distributed across versions, with customers in the same code receiving the same version. The authors describe it as approximately random and checked demand on other catalog items. That is a useful protection against differences between audiences, but it remains more specific than a perfectly independent visitor level website experiment. (Anderson and Simester, May 2001)

How should the demand indices be read?

The five newly labeled products had a treatment index of 157 against their control index of 100. The three already labeled products had a treatment index of 38 against their own control index of 100. Both differences were significant at the .05 level. The accessory's index was 90 against 100, without a significant reported difference. (Anderson and Simester, May 2001)

The arithmetic changes are positive 57% and negative 62% for the two product groups. Each percentage uses that group's control demand. A group with a smaller starting quantity can have a large percentage change while contributing fewer units to total demand. The download therefore preserves the index unit rather than inventing raw sales counts. (Anderson and Simester, May 2001)

Do not add 157 and 38, subtract two 100 baselines and call the difference a net category effect. The source explicitly warns that the table cannot be used for that comparison. It separately states that aggregate demand across the eight products did not differ significantly. (Anderson and Simester, May 2001)

Does the test prove that sale labels lost credibility?

It is consistent with a credibility explanation, but the field test alone also allows substitution. The added labels may have redirected shoppers toward the five test products and away from the previously labeled three. The paper says that this particular experiment cannot discriminate between those explanations. (Anderson and Simester, May 2001)

The article uses additional historical catalog data, grocery scanner data and a survey to investigate the larger argument. Those components have different designs. A retailer should not present every supporting result as if the same randomized test proved an optimal label share across categories.

This distinction changes implementation. If the problem is substitution, evaluate the assortment's contribution and which products gain or lose. If shoppers no longer trust promotion signals, evaluate their understanding of actual prices and conditions. Both concerns matter, but an item level lift alone does not tell the team which explanation applies.

What should a store measure before adding more badges?

Measure the entire eligible assortment, including products that keep their existing labels. Preserve prices and promotional conditions if the goal is to isolate presentation. Otherwise a badge coverage test can become a combined price, availability and messaging experiment.

Track category revenue, contribution and unit mix, with the same assignment across the shopper's journey. A cart recovery message should reflect the same valid price and offer that the customer sees on returning. A label whose offer no longer exists creates an answer reliability problem as well as a promotion problem.

The source compares 37.5% and 100% label coverage among eight focal products. It does not identify a universal optimum between those values. A store's decision should follow its own assortment, substantiated sale conditions and aggregate results, while retaining the possibility that a visible gain on the newly promoted items is offset elsewhere. (Anderson and Simester, May 2001)

What baseline mix would make the reported changes balance?

The unreported group weights prevent a net-demand calculation from Table 3 alone. They do permit a conditional accounting question: what weights would be needed for the newly labeled products' gain to offset the other changes?

Let the three baseline quantities be the units sold by the newly labeled group, the previously labeled group and the accessory. Applying the reported point indices gives a change of 57% of the first quantity, minus 62% of the second, minus 10% of the third. Those quantities are group totals, not the counts of products within the groups. (Anderson and Simester, Table 3, May 2001)

Ignoring the accessory, the newly labeled group's baseline unit demand would have to exceed approximately 1.088 times the previously labeled group's baseline demand to offset these point changes. This is our conditional threshold, calculated as .62 divided by .57. It is not an estimate of the actual category result. The source does not supply the necessary group volumes in this table, and the accessory's separate 10% decline was not statistically significant.

Conditional demand accounting from the published point indices: add 0.57 times the newly labeled group's baseline units, subtract 0.62 times previously labeled baseline units and subtract 0.10 times accessory baseline units. Actual baseline group weights are unknown from Table 3.
Figure 2: The baseline sales mix determines the sign of the weighted sum. Counting five products against three cannot supply the missing demand weights. Source: Anderson and Simester, Table 3, May 2001. Original comparison from the cited study. Study period and calculation limits appear in the text.

For a simple worked comparison excluding accessories, suppose both groups initially sold 100 units. Applying the reported relative changes gives a gain of 57 units and a loss of 62, a net decline of five. If the newly labeled group instead began at 200 units and the other at 100, the same relative changes give a net gain of 52 units. These are declared example weights. Neither scenario reconstructs the retailer's actual sales.

The exercise explains why an impressive product-group percentage does not answer the assortment decision. It also leaves the paper's actual conclusion intact: the authors separately report no significant difference in aggregate demand across the eight dresses. Nonsignificance does not prove an exactly zero effect, and a conditional accounting threshold does not overturn the source's aggregate test.

Why might the contribution result differ from the unit result?

Unit weights are only the first layer. A merchant ultimately needs the contribution associated with each product group's demand. If the newly promoted products have different margins, a unit increase can offset an assortment unit decline yet still reduce contribution, or the reverse. Table 3's indices do not provide the group-level prices and costs needed to calculate that result.

For a current test, retain untreated baseline volumes, actual treatment volumes and variable contribution for every affected group, including products whose labels do not change. A promotion can divert demand away from an existing bestseller. Dropping that product from the report because it was not newly labeled would remove part of the treatment's business consequence.

Keep assortment-wide outcomes separate from explanations about why a customer switched. The historical field test allows both credibility and substitution mechanisms. The weighted accounting tells the merchant where gains and losses occur. It cannot determine what shoppers believed about the labels, and it cannot establish that the store should remove all sale cues.

Which calculations can a reader reproduce?

The download retains the published inputs, their units and their comparison populations. The following arithmetic uses those inputs. It does not add a new sample or a new causal comparison.

CalculationExpressionResult
Newly labeled item demand change, percent(157/100-1)*10057.000
Previously labeled item demand change, percent(38/100-1)*100-62.000
Control focal label coverage, percent3/8*10037.500
Treatment focal label coverage, percent8/8*100100.000

Methodology and Sources

We read the cited primary source and the decisive methods, results and exceptions on 13 November 2026, retaining its PDF and text under the source ledger. The download distinguishes reported aggregates from our arithmetic, with units and analysis populations preserved. The formulas shown here can be rerun from analyze.py. We do not reconstruct individual records from rounded aggregates or treat an implementation suggestion as a measured result. (Anderson and Simester, May 2001).

Each row has its own 100 baseline, so rows cannot be summed to estimate category sales. The paper reports no significant aggregate demand difference across all eight products. Approximate geographical randomization and one seasonal catalog constrain transfer. The design cannot separate credibility from substitution by itself and tests only two coverage levels.

Glossary

  • Demand index: A normalized measure that sets the comparison group to 100 within each row.
  • Own control baseline: The comparison quantity for that specific group of products, not another row’s quantity.
  • Substitution: Purchases shifting from one item to another.
  • Aggregate demand: Total purchases across the specified group of products.
  • Label coverage: The share of the studied assortment carrying a sale label.
  • Assignment unit: The unit allocated to a treatment condition. In this test, customers within a zip code received the same catalog version.

Sources

Cite this page

Use this citation when referring to the analysis. Retain the original study scope with any figures.

Said Maybe. "Sale label saturation statistics and demand shifted between products." Said Maybe, November 2026. https://saidmaybe.com/guides/sale-label-saturation-product-demand-statistics/
<a href="https://saidmaybe.com/guides/sale-label-saturation-product-demand-statistics/">Sale label saturation statistics and demand shifted between products</a> (Said Maybe, November 2026)