More loans can accompany fewer reported users
An increase in digital-library loans does not necessarily mean that the service reached more annual users. In Parma's published EMILIB series, reported ebook loans increased by 1,020 between 2021 and 2022 while the reported unique ebook-user count fell by 168. Both statements come from the same official annual table. They are not contradictory: a total and the number of users associated with that service measure different aspects of its reported activity. Reading only the total would miss the opposing movement.
The longer comparison is different again. From 2018 to 2022, cumulative registrations almost tripled, the unique ebook-user count grew by roughly two thirds, and ebook loans slightly more than doubled. These are three administrative count paths, not three estimates of the same population. Registrations since the beginning of a service accumulate across time, whereas a unique-user count belongs to a particular year. Their different growth does not by itself identify inactive current members, lost readers or a failure to convert registrations into borrowing.
This study uses all five published years to ask a bounded question: how did these reported counts move together, and how can changes in the loan total be separated into user-count and loans-per-reported-user accounting components? The answer preserves the years in which those components point in different directions. It does not reconstruct individual behavior. Its practical contribution is to show why a registration milestone or a rising transaction total cannot stand alone as a description of the breadth of annual ebook use.
The Parma reporting footprint, not the whole platform
The source is the Comune di Parma open-data dataset covering use of the EMILIB MLOL digital library from 2018 through 2022. Its description refers to residents of the provincial territory, registered mainly through Parma municipal libraries but also through other local-authority libraries and the Palatina. It additionally identifies a significant university-student component. This is the publisher's reporting footprint. It is not a census of every Parma resident, a national Italian series or the total activity of every institution participating in EMILIB.
The distinction is important because the regional digital-library platform spans a larger territory. Official regional commentary describes EMILIB as serving six provinces and distinguishes it from MLOL Romagna. Parma's records cannot simply be added to another municipality's figures without knowing whether the populations, registration systems and transactions overlap. Nor can a Bologna-labelled series supply missing Parma definitions. The present analysis uses one jointly published historical series and does not splice in figures from another service report to lengthen it.
Publishing five years together is evidence of an intended reporting series, not proof that every underlying convention remained immutable. The inspected metadata do not fully document annual changes in participating institutions, eligibility, account handling or loan conventions. That uncertainty limits interpretation throughout the article. Changes are therefore described as movements in the published counters. They are not automatically changes among a fixed set of people, and they cannot isolate the effect of library policy, the pandemic, university enrollment or any other cause.
What each counter does and does not measure
The source distinguishes new registrations from registrations since inception. The latter is the cumulative counter used here, denoted R. It is not independently verified as a register of currently eligible, living or otherwise active members. The unique ebook-user count, denoted B, is the source-labelled annual measure for ebook users. The label does not establish that each counted user is a distinct person, that every user completed a loan, or that the same credentials and deduplication rules operated unchanged throughout the period.
Reported ebook loans, denoted L, are transaction counts, not books completed or titles read. The available dataset does not establish how renewals or repeat borrowing are handled in this particular counter. The quotient L divided by B is consequently named loans per reported ebook user. It is a division of jointly published operational measures. Calling it individual reading intensity would add a behavioral interpretation that the source does not verify. In particular, an increase in this quotient does not prove that identified continuing users read more.
The source also publishes overall unique users, unique newsstand users, total accesses and newsstand consultations. All remain available in the accompanying input file, but they are not combined into a synthetic readership total. Ebook and newsstand users can overlap, and the relationship between the overall count and each category is not fully documented. The analysis therefore does not subtract category users to invent nonusers, sum unique-user categories, calculate a platform-to-ebook conversion rate, or equate an access with a reading session.
Five years and a transparent registration check
The complete published series contains eight metrics and five annual columns, with no missing numeric cell in those forty source values. All years from 2018 to 2022 were fixed before derived outcomes were calculated. The primary endpoint comparison uses 2018 and 2022, and the analysis also retains every adjacent annual transition. This prevents an attractive endpoint from replacing the intermediate observations that make the result more informative. The final source year is 2022; the article makes no claim about activity in later years.
One useful arithmetic check concerns registrations. For each transition from 2019 through 2022, the change in registrations since inception equals that year's published new registrations. The cumulative values are 7,207, 10,098, 17,023, 19,133 and 21,336; the corresponding additions after 2018 are 2,891, 6,925, 2,110 and 2,203. This exact reconciliation supports reading the two fields as an accumulated counter and its recorded additions. It does not prove that no account was deleted, replaced or otherwise reclassified behind the published totals.
For that reason, the study does not divide annual ebook users by cumulative registrations and call the result an activation rate. A defensible rate of current-member activation would require a compatible currently eligible membership denominator. Instead, the main comparison places each counter on its own 2018-equals-100 scale. That shows relative movement without pretending that historical registrations define a current population at risk. Absolute annual counts remain alongside the indices so that a common graphical baseline does not conceal different units or starting values.
Registration, user and loan paths tell different stories
Between 2018 and 2022, registrations since inception rose from 7,207 to 21,336, a growth factor of 2.96. Reported unique ebook users rose from 1,953 to 3,238, a factor of 1.66. Ebook loans rose from 16,752 to 35,168, a factor of 2.10. The cumulative registration index therefore reaches 296.05, the ebook-user index 165.80 and the loan index 209.93 when 2018 is set to 100. None of these differences requires an assumption about which individuals returned.
The annual user path is not steadily increasing. Reported unique ebook users rise to 2,524 in 2019 and 4,262 in 2020, then fall to 3,406 in 2021 and 3,238 in 2022. Loans reach 22,891 in 2019 and 37,535 in 2020, decline to 34,148 in 2021, then increase to 35,168 in 2022. The last transition thus moves in opposite directions for users and loans. The accumulated registration counter continues upward, as expected from its published positive additions, but it cannot resolve that divergence.
A service could accurately report a larger loan total and still leave readers uncertain about the breadth of its annual ebook use. Conversely, a falling user count does not imply that every aspect of activity declined. These are reasons to present the counters together, not to choose whichever produces the most favorable headline. They also explain why the comparison is descriptive rather than evaluative: the data alone do not establish service quality, equitable access, reader satisfaction, unmet demand or the reasons behind the changes.
An exact accounting bridge, not a causal explanation
Define I as L divided by B for each year. Then the identity L equals B times I holds by construction. To compare two years, this study uses a symmetric decomposition. The user-count component is the change in B multiplied by the average of the two I values. The quotient component is the change in I multiplied by the average of the two B values. Adding the components recovers the observed loan-count change exactly, without leaving an unexplained interaction remainder.
Symmetry matters because changing one factor first and the other second would otherwise make the allocation depend on the chosen order. Averaging the two possible orders gives the stated formula. This is an established accounting approach, not a new statistical theory. It answers how a change in the product can be represented using changes in its two reported factors. It does not establish which factor caused the other, assign transactions to particular users, or separate new, returning and continuing accounts.
The components can be negative, and their displayed values can contain decimals even though observed loans are integers. Those decimals are accounting equivalents produced by the formula, not fractional transactions found in the source. All calculations use exact rational arithmetic before rounding for presentation. The five comparisons include the primary 2018-to-2022 endpoints and all four adjacent transitions. No confidence intervals, significance tests or fitted time trends are attached to these five administrative observations, and no component is converted into a causal percentage contribution.
The endpoint bridge and the first two transitions
Across the full period, the observed loan increase is 18,416. Its symmetric user-count component is 12,489.30 and its loans-per-reported-user quotient component is 5,926.70 accounting equivalents. Both components are positive. This means that both reported factors increase between the selected endpoints. It does not mean that 12,489 particular loans were made by newly recruited people or that 5,927 loans were additional reading by a known continuing group. Such statements would require linked records that are absent here.
The first annual transition, 2018 to 2019, also has two positive components. The loan increase of 6,139 is represented by 5,038.19 from the user-count change and 1,100.81 from the quotient change. Showing this transition establishes that the complete series contains more than a single exceptional year. It also keeps the accounting interpretation consistent: the components describe the same published counters in every period, rather than switching definitions to tell a smoother narrative about growth.
The next transition provides a different pattern. From 2019 to 2020, loans increase by 14,644, but the quotient component is negative, at −890.45. The user-count component is positive and larger than the net change, at 15,534.45. A growing total can therefore coexist with a declining loans-per-reported-user quotient. The calendar timing overlaps the pandemic, but these records do not identify a causal pandemic effect or distinguish restrictions, acquisitions, eligibility, account changes and other possible explanations. The signed decomposition is the supported result.
The later transitions expose opposing movements
From 2020 to 2021, loans decline by 3,387. The user-count accounting component is −8,060.41, while the quotient component is +4,673.41. The net decline is smaller than the negative user-count component because the second component points in the opposite direction. This representation is more informative than calling the entire transition a uniform contraction. Yet it remains an identity among operational counters, not evidence that a specific group compensated for departing users or that remaining members changed their habits.
From 2021 to 2022, the reported unique ebook-user count falls from 3,406 to 3,238, while loans rise from 34,148 to 35,168. The corresponding components are −1,754.50 and +2,774.50, summing to the observed increase of 1,020. This is the clearest example of why loan totals and user-count breadth should not be used interchangeably. A positive total can coexist with fewer source-labelled annual ebook users, even though the cumulative registration counter has also continued to rise.
The endpoint and annual bridges answer related but different questions. Each is exact for its own interval, but the separate user-count components from annual bridges need not sum to the endpoint user-count component. The symmetric formula uses the factor levels at each interval's endpoints, so its allocation depends on the comparison window. The total loan changes do telescope across years. Readers should therefore compare the displayed annual sign patterns and the endpoint summary without treating their allocated components as additive historical cohorts.
What this adds beyond a digital-growth headline
Official regional commentary already reported increases in digital-library access and ebook loans during the early pandemic period. The existence of expanding digital services is not a new finding, and the Parma source itself already publishes the annual levels. The contribution here is narrower: separating an accumulated registration count from annual user counts, retaining the entire five-year path, and showing exact signed accounting bridges when the loan total and its reported factors move differently. This is not another league table of libraries or regions.
That approach provides a practical reading rule for administrative headlines. Ask whether a number accumulates since launch or refers to one year; whether users are persons, accounts or an operational category; and whether a transaction total can change while its associated user count moves the other way. In this series the final question has a concrete answer, not merely a generic caveat. The rule is useful without declaring the service a success or a failure and without claiming to measure how much reading people completed.
The analysis also avoids turning unknown overlap into an apparent discovery. It would be tempting to combine ebook and newsstand users, label the difference from overall users as disengagement, or interpret cumulative registrations without annual ebook use as dormant members. None is established by these aggregate fields. The source supports the reported count comparisons and their arithmetic identities. It does not supply the account-level links or current eligibility histories needed for retention, cross-format behavior or individual transitions. Keeping that boundary explicit protects the useful result from a stronger unsupported interpretation.
Reproduction and the limits of the conclusion
The public reproduction package contains the complete forty-cell source table in a normalized UTF-8 file, the original source fingerprint and URLs, calculation methods and a standard-library Python program. The program accepts explicit input and output paths and reproduces the registration checks, annual quotients and indices, growth factors and every decomposition. It neither contacts users nor requests private library records. Original source bytes are retained separately for provenance; the normalized file changes line endings only, not labels or values. The dataset carries a Creative Commons Attribution 4.0 licence.
Several uncertainties remain outside the arithmetic. The source does not establish unchanged account or institutional coverage, explain every deduplication convention, document loan renewals for this table, or prove that cumulative registrations remain valid memberships. The user measure must retain its source label rather than becoming a verified borrower or person count. Five annual observations are insufficient for causal inference about policy or the pandemic, and the series ends in 2022. Independent numerical agreement establishes reproducibility, not an answer to those source-definition questions.
The defensible conclusion is consequently specific. In Parma's published 2018–2022 digital-library counters, cumulative registrations, annual unique ebook users and ebook loans grow at different rates; in 2021–2022, loans increase while the user count declines. Exact signed accounting components make those contrasts visible without reconstructing individual behavior. Registration milestones and lending totals are useful descriptions, but neither replaces the annual user-count measure. Presenting all three, with their distinct time bases and documented limits, gives a fuller account of the reported service activity than any single growth number.
Registration, ebook-user and loan counts follow different paths
Parma reporting footprint · EMILIB · All five years, 2018–2022
Full chart and source
Scroll horizontally to inspect the chart. ↔
Chart and values
Scroll horizontally to see all columns.
| Year | Cumulative registrations index | Unique ebook users index | Ebook loans index |
|---|---|---|---|
| 2018 | 100.00 | 100.00 | 100.00 |
| 2019 | 140.11 | 129.24 | 136.65 |
| 2020 | 236.20 | 218.23 | 224.06 |
| 2021 | 265.48 | 174.40 | 203.84 |
| 2022 | 296.05 | 165.80 | 209.93 |
Each counter has its own 2018 baseline. Common index scale does not establish a common population denominator or immutable scope.
Loan changes contain opposing signed accounting components
Parma reporting footprint · EMILIB · All five years, 2018–2022
Full chart and source
Scroll horizontally to inspect the chart. ↔
Chart and values
Scroll horizontally to see all columns.
| Comparison | Observed loan-count change | User-count component | Quotient component | Exact residual |
|---|---|---|---|---|
| 2018–2022 | 18,416 | 12,489.30 | 5,926.70 | 0 |
| 2018–2019 | 6,139 | 5,038.19 | 1,100.81 | 0 |
| 2019–2020 | 14,644 | 15,534.45 | −890.45 | 0 |
| 2020–2021 | −3,387 | −8,060.41 | 4,673.41 | 0 |
| 2021–2022 | 1,020 | −1,754.50 | 2,774.50 | 0 |
User-count plus quotient component equals observed loan-count change exactly before rounding. Fractional contributions are accounting equivalents, not observed fractional transactions. Endpoint allocations need not equal sums of annual allocations.
Data & downloads
Scroll horizontally to see all columns.
| Year | Registrations since inception | New registrations | Unique ebook users | Ebook loans | Loans per reported ebook user |
|---|---|---|---|---|---|
| 2018 | 7,207 | 2,767 | 1,953 | 16,752 | 8.58 |
| 2019 | 10,098 | 2,891 | 2,524 | 22,891 | 9.07 |
| 2020 | 17,023 | 6,925 | 4,262 | 37,535 | 8.81 |
| 2021 | 19,133 | 2,110 | 3,406 | 34,148 | 10.03 |
| 2022 | 21,336 | 2,203 | 3,238 | 35,168 | 10.86 |
Source operational counters. Cumulative registrations are not verified current memberships; ebook users are not necessarily persons or borrowers. Quotient L/B is not individual reading intensity.
Scroll horizontally to see all columns.
| Year | Cumulative registrations index | Unique ebook users index | Ebook loans index |
|---|---|---|---|
| 2018 | 100.00 | 100.00 | 100.00 |
| 2019 | 140.11 | 129.24 | 136.65 |
| 2020 | 236.20 | 218.23 | 224.06 |
| 2021 | 265.48 | 174.40 | 203.84 |
| 2022 | 296.05 | 165.80 | 209.93 |
Each counter has its own 2018 baseline. Common index scale does not establish a common population denominator or immutable scope.
Scroll horizontally to see all columns.
| Comparison | Observed loan-count change | User-count component | Quotient component | Exact residual |
|---|---|---|---|---|
| 2018–2022 | 18,416 | 12,489.30 | 5,926.70 | 0 |
| 2018–2019 | 6,139 | 5,038.19 | 1,100.81 | 0 |
| 2019–2020 | 14,644 | 15,534.45 | −890.45 | 0 |
| 2020–2021 | −3,387 | −8,060.41 | 4,673.41 | 0 |
| 2021–2022 | 1,020 | −1,754.50 | 2,774.50 | 0 |
User-count plus quotient component equals observed loan-count change exactly before rounding. Fractional contributions are accounting equivalents, not observed fractional transactions. Endpoint allocations need not equal sums of annual allocations.
Evidence & method
How we calculated it
- Exploratory secondary analysis of one complete official Parma 2018–2022 reporting series; all five years and every adjacent transition fixed before derived outcomes.
- Preserve all 40 source cells. R denotes registrations since inception, B source-labelled unique ebook users, L reported ebook loans, and A the source-labelled overall unique users. No B/R, B/A or A−B rates.
- Exact rational symmetric decomposition: ΔB times average(L/B), plus Δ(L/B) times average B, equals ΔL. Components are signed accounting equivalents, not causal effects.
- Annual 2018=100 indices and fixed 2018→2022 growth factors; registration difference-versus-new-registration audit. No confidence intervals or fitted trends.
- Public standard-library CLI reproduction with explicit paths, source hashes and CC BY 4.0 attribution.
What the data cannot tell us
- Common published historical series does not certify immutable institutions, eligibility, account deduplication or loan conventions.
- Unique ebook users are not independently verified distinct people or borrowers; loans-per-reported-user is an operational quotient, not individual reading intensity.
- Registrations since inception are not verified currently valid memberships; no activation, retention or dormant-person claims.
- Ebook/newsstand overlap and overall-user nesting are not inferred. Loan renewals and completed reading are not identified.
- Five administrative years ending 2022; no causal pandemic, service quality, population or later-year inference.
Original sources
- Comune di Parma: EMILIB MLOL users and loans, 2018–2022
Official jointly published historical series; provincial-resident/library and university-student footprint. Explicitly links CC BY 4.0. Reporting scope is not verified immutable people or accounts.
Accessed October 30, 2026
- Comune di Parma: original EMILIB CSV
Eight source-labelled metrics across five complete years. New registrations and registrations since inception are separate fields; unique ebook users are not independently verified borrowers or persons.
Accessed October 30, 2026
- Comune di Parma: dataset resource metadata
Official 440-byte CSV resource; timestamp 24 January 2023. Dataset landing modified 13 March 2023. These are source publication metadata, not observations beyond 2022.
Accessed October 30, 2026
- Regione Emilia-Romagna: digital-library activity during March 2020
Prior official digital-growth commentary and six-province EMILIB versus MLOL Romagna scope. Not a compatible Parma denominator or source of causal identification.
Accessed October 30, 2026
- Creative Commons Attribution 4.0 International
Licence explicitly linked by the official Parma dataset. Credit Comune di Parma/Sistema Bibliotecario; identify normalized inputs and new calculations. No official endorsement implied.
Accessed October 30, 2026
Cite & reuse this study
Alisson Bento. When digital-library loans rise but reported users fall. Secondary analysis of Comune di Parma EMILIB reporting counts, 2018–2022. Sources accessed October 30, 2026. https://alissonbento.com/research/parma-digital-library-registration-users-and-loans-2018-2022/
Source numeric data: Comune di Parma, Sistema Bibliotecario, CC BY 4.0. Normalized input labels and values retained; original calculations and explanation by Alisson Bento. No official endorsement, logo or third-party report artwork.
Corrections or editorial questions