Skip to content

Brazil · Consumer-unit budgets · POF 2017–2018

The income gap in Brazilian non-textbook acquisition: incidence and conditional value

The incidence component accounts for 66.59% of the highest–lowest income-band gap in mean 90-day monetary non-textbook acquisition value under a symmetric decomposition. The result describes officially processed records, including imputation, not a causal income effect or annual buyer probability.

Published
Share
Share this page
EmailCite & reuse this study
Key findings as text
Reusable imagesThe income gap in Brazilian non-textbook acquisition: incidence and conditional value — 66.59%: Incidence component share of the highest–lowest mean-value gap. Symmetric accounting decomposition of processed 90-day monetary acquisition, not causal attribution.Download sharing image (JPG) ↓

Key findings

66.59%

Incidence component share of the highest–lowest mean-value gap

Symmetric accounting decomposition of processed 90-day monetary acquisition, not causal attribution.

0.77% → 23.80%

Lowest to highest income-band 90-day processed acquisition incidence

Monthly total CU income includes asset variation; monetary modes 1–6, not annual buyers.

R$39.08

Highest-minus-lowest mean 90-day acquisition-value gap per CU

Historical 15 January 2018 price reference; not cash outflow or current reais.

A larger incidence component in a historical income gap

The difference between the highest and lowest income bands in Brazil's 2017–2018 household budget survey is associated with both more widespread monetary book acquisition and higher acquisition value among the consumer units with a positive record. Under the symmetric accounting decomposition used here, the incidence component represents 66.59% of the difference in mean ninety-day non-textbook acquisition value. The conditional-value component represents the remainder. This is a description of the processed survey records, not an estimate of what would happen if a family's income increased.

The comparison uses the highest and lowest of seven official bands of monthly total consumer-unit income, including asset variation. It is not a comparison of disposable earnings, individual salaries or wealth groups. Within the lowest band, processed positive monetary acquisition incidence is 0.77%; within the highest it is 23.80%. Conditional ninety-day acquisition value is R$59.86 and R$166.16 respectively. The combination produces a mean-value gap of R$39.08 per consumer unit over the ninety-day reference window, expressed at the survey's historical price reference.

These figures answer a narrower question than whether poorer people read. The survey does not observe every way of obtaining or reading a book. Nor does a ninety-day absence of monetary acquisition establish that a consumer unit never buys books. The contribution of the analysis is to separate two components that an unconditional mean combines, while retaining the boundaries imposed by the questionnaire, the official processing and the reference period.

The unit is a consumer unit, not an individual reader

The POF defines a consumer unit through shared food provision and, where that criterion does not apply, housing expenses. A consumer unit can consist of one resident or several residents. A single household can contain more than one consumer unit. This distinction matters both for the denominator and for the meaning of the income bands. The analysis contains 58,039 consumer units in 57,920 households. Neither count is the number of book readers in the sample.

The resident file establishes the complete consumer-unit frame before book-acquisition records are added. Each consumer unit enters the denominator once, including those without any qualifying book record. Its repeated resident-level final weight and income are checked for agreement and retained once. Summing those repeated weights across residents would instead weight larger units several times and would not reproduce a consumer-unit statistic. Likewise, using only the expense file would omit units without the selected acquisitions and inflate incidence.

The weighted consumer-unit total is approximately 69.02 million. The overall total and all seven income-band totals match the later official published counts after rounding. This is a useful frame check, not proof that every underlying acquisition was recalled correctly. All expense records also link to the resident frame, and identifiers and repeated weights are internally consistent. These checks establish the denominator and joins; they do not turn household survey responses into a complete transaction ledger.

What counts as a non-textbook acquisition

The primary category consists of printed and digital non-textbooks in questionnaire block 32. The printed product family includes the general non-textbook label and synonyms for printed Bibles, catechisms, fascicles and novels. The digital family identifies digital non-textbooks. The manual describes recreational, informative, religious and self-help reading materials that were not acquired as teaching support or an integral part of a course or school activity. This is therefore neither a fiction-only category nor a measure of recreational reading alone.

The analysis excludes school and technical books classified elsewhere, newspapers, magazines, periodical subscriptions and coloring books. That deliberate boundary is narrower than the official general expenditure group combining periodicals, books and non-didactic magazines. Substituting the broad aggregate would introduce other products and reference periods. It could not answer the same two-part question simply by relabeling the result as book expenditure. The two selected product prefixes are retained consistently across the primary calculation and the own-unit sensitivity.

Only monetary acquisition modes enter the main calculation: purchases for the consumer unit itself or for another consumer unit, whether recorded as immediate payment, installment purchase or credit-card acquisition. Donation receipts, business withdrawals, barter, own production and the residual other mode are excluded. The resulting value is not cash paid during the window, because acquisition and payment schedules need not coincide. It also is not the value of every book received by the unit, since nonmonetary receipts are outside the primary measure.

Ninety-day incidence must stay ninety-day incidence

Block 32 refers to acquisitions during ninety days. For each consumer unit, the analysis sums the valid, corrected monetary values of the selected records and identifies whether that sum is positive. Incidence is the weighted proportion of all consumer units in the income band with a positive processed sum. The conditional value is the weighted mean of that same ninety-day sum among those positive units. The unconditional mean includes every consumer unit, assigning zero when the final data contain no qualifying record.

This common time base makes the identity transparent: incidence multiplied by conditional value equals unconditional mean value. It would be incorrect to combine an annual buyer probability with a conditional amount defined for ninety-day acquirers, because the populations and reference windows would differ. In particular, multiplying the incidence by four does not produce annual buyer incidence. A unit may acquire books in several quarters, in one quarter only, or in a pattern not recoverable from this cross-sectional survey.

The official annualization factor for these records is four. Multiplying acquisition value by four therefore gives an annualized intensity consistent with the survey's expenditure convention. It does not reveal each unit's actual annual purchases. Even an annualized conditional value remains conditional on a positive ninety-day record. To keep this distinction visible, the article and charts use ninety-day values as the main presentation. The reproduction script also calculates annualized values with explicit labels, without estimating annual participation.

Reading the full seven-band profile

The complete profile shows why both parts of the mean matter. Processed incidence rises across the seven ordered income bands from 0.77% to 1.56%, 3.13%, 7.78%, 10.91%, 15.85% and 23.80%. These percentages refer to consumer units within each band, not to each band's share of all acquisitions or all readers. The bands are not equal-sized quintiles. Their population counts differ considerably, so an incidence percentage alone does not describe how many acquisitions a band contributes nationally.

Conditional ninety-day values do not rise at every adjacent step. The first two bands have values of R$59.86 and R$56.67. Subsequent bands have R$94.92, R$95.91, R$102.15, R$118.28 and R$166.16. These are point estimates, and the article does not claim that every adjacent contrast is statistically distinguishable. The lower conditional value in the second band is a reminder that the full profile is more informative than an assumed smooth relationship between income and purchases.

Each band contains at least 130 positive primary consumer units in the achieved sample. The positive sample counts are not weighted population totals and should not be used as the denominator for the displayed incidence percentages. The accompanying table reports the sample bases separately. The national scope avoids multiplying the analysis into potentially sparse geographic cells. Regional or state-level findings are not implied by the availability of national estimates, and this study does not rank places or extrapolate the pattern to particular municipalities.

A symmetric accounting decomposition, not a policy experiment

The highest-minus-lowest mean gap is R$39.08 over ninety days. The decomposition assigns R$26.02 to the incidence component and R$13.06 to the conditional-value component. The incidence term multiplies the difference in incidence by the average of the two conditional values. The value term multiplies the difference in conditional values by the average of the two incidence rates. Their sum equals the total mean gap, apart from displayed rounding. Using full-precision values preserves the identity exactly to numerical tolerance.

The symmetric form avoids choosing one income band's level as the sole reference for both changes. It averages the two possible orderings of the accounting exercise. Nevertheless, it is still a particular descriptive decomposition rather than a unique causal attribution. Its 66.59% incidence share should not be read as the proportion of inequality that an access program would remove. Nor does it establish that changing the number of acquirers would leave conditional values unchanged in a real intervention.

Consumer units in the bands differ in composition, education, ages, preferences, access to shops and libraries, and many other characteristics. The study does not adjust those differences away or follow the same units through an income change. Its estimand is intentionally the observed cross-sectional contrast in the officially processed data. An explanation of why the components differ would require additional evidence and a different design. The present result supplies a clear descriptive starting point without claiming to identify a mechanism.

Income bands and reais have historical definitions

The income variable is monthly gross total consumer-unit income, including monetary and nonmonetary income and asset variation as defined by IBGE. Asset variation can include property sales, inheritances and a positive balance of financial movements. Those items are important qualifications, not incidental footnotes. A band based on this variable is not a stable earnings class, a measure of disposable resources alone, or a classification of accumulated wealth. Consumer units with the same recorded total can have very different sources and regularity of resources.

The six thresholds are R$1,908, R$2,862, R$5,724, R$9,540, R$14,310 and R$23,850 per month. The lowest band includes units without income. The classification uses the corrected income variable in the official microdata. IBGE announced a correction to that derived variable in October 2019 and an update of income-class tables in November 2019. The original first-results publication therefore cannot be used uncritically as the exact band-count benchmark, even though its overall consumer-unit total agrees.

The monetary values use the official corrected acquisition amounts with a reference price date of January 15, 2018. They are not today's reais and should not be compared directly with a current book price, household budget or benefit amount. No contemporary affordability threshold is estimated. Likewise, the historical bands should not be translated into present-day income groups by simply keeping their nominal cutoffs. The analysis retains the source definitions to answer a historical question, rather than presenting old amounts as a current market description.

Official imputation changes how incidence should be described

The final microdata are processed survey data. IBGE describes hot-deck procedures for missing or rejected values and for missing parts of questionnaires or whole questionnaires. This means that a positive acquisition record need not represent a directly reported transaction in every case. The main estimates retain the official corrected values, as required for consistency with the published survey framework. The language used here is consequently processed monetary acquisition incidence, not a clean count of directly observed buying events.

The narrow category contains 2,117 monetary records. Twenty-two carry the explicit amount-imputation flag. However, 252 have a blank deflator, a condition the dictionary associates with block-imputed values that are already price-corrected. Those records all have the amount-imputation flag set to zero. The blank-deflator diagnostic affects 241 consumer units. These distinctions show why an apparently simple filter on the amount flag would not isolate an entirely nonimputed sample or eliminate uncertainty about acquisition participation.

Twenty-three monetary records have an unknown raw amount but a valid corrected value. Unknown raw amounts are not treated as zeros or as their numeric sentinel. The analysis uses the official corrected amount and does not apply a second deflation to values already corrected. Missing income or an invalid join would also not be recoded as a genuine zero. Careful handling of these fields prevents mechanical errors, but it does not quantify the uncertainty introduced by the imputation process or recover the missing original answers.

Sensitivities retain the measurement limits

The main monetary definition includes purchases for another consumer unit. A separate sensitivity limits the selection to monetary acquisition for the unit itself. Under that restriction, the incidence component accounts for 66.39% of the highest-minus-lowest mean gap, compared with 66.59% in the primary analysis. This limited difference indicates that the descriptive split is not driven principally by the inclusion of monetary acquisitions made for other units under the recorded acquisition-mode classification.

A second, separately documented diagnostic excludes records with the explicit amount-imputation flag. Its incidence share is 66.44%. This is not an observed-only estimate. The blank-deflator evidence demonstrates that whole-block processing is not captured completely by that flag. The closeness of these point results should therefore not be advertised as proof that imputation is unimportant or that alternative treatments of missing questionnaires would leave the result unchanged. It establishes only what the specified exclusion changes in the available processed records.

Sampling uncertainty remains relevant as well. Independent implementations agree on an approximate Taylor primary-cluster calculation, but exact finite-population correction and calibration treatment have not been established for an official-equivalent variance estimate. Additional imputation and recall uncertainty are not included in that calculation. The main findings therefore give point estimates without confidence intervals, significance claims or assertions that every adjacent income band differs. Agreement between two implementations verifies the computation, not every assumption needed for comprehensive statistical inference.

What the study contributes and what it leaves open

A mean expenditure gradient can arise because more units record an acquisition, because the positive units record a higher value, or because both occur together. In this historical Brazilian non-textbook category, the symmetric decomposition assigns the larger point contribution to processed incidence. That conclusion is more specific than saying that higher-income units spend more, but less expansive than a theory of reading demand. It does not establish willingness to read, the number of books consumed, or the effect of prices on participation.

Zero monetary acquisition can coexist with reading books already owned, borrowing from libraries or other people, receiving gifts, obtaining free material, or acquiring books outside the ninety-day window. A positive acquisition can also be a purchase for someone else. The data do not connect a particular purchase to its eventual reader or establish whether a book was read at all. Reading culture, publishing revenue and household acquisition value are related topics, but they are not interchangeable outcomes in this study.

The public reproduction materials contain only national income-band aggregates and original calculation code. They allow the point estimates and decomposition to be regenerated without exposing household, person or primary-sampling-unit identifiers. An optional local check accepts exact official source files supplied by the reader, verifies their hashes and recomputes the national totals without exporting individual records. This separation preserves a useful audit trail while avoiding the publication of raw microdata or sparse sampling-unit cells merely to make a statistical calculation convenient.

Two parts of monetary non-textbook acquisition

Brazil, POF 2017–2018; seven monthly total CU income bands including asset variation. Processed ninety-day incidence and conditional acquisition value; historical price basis.

Full chart and source
Seven income bands: incidence 0.77, 1.56, 3.13, 7.78, 10.91, 15.85, 23.80 percent; conditional 90-day values 59.86, 56.67, 94.92, 95.91, 102.15, 118.28, 166.16 reais.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

All seven income bands: processed monetary non-textbook acquisition
Monthly CU income (R$)90-day incidence (%)Conditional 90-day value (R$)Mean 90-day value (R$)Sample CUsPositive CUs
≤ 1,9080.7759.860.4616,583130
> 1,908–2,8621.5656.670.8811,839154
> 2,862–5,7243.1394.922.9817,246470
> 5,724–9,5407.7895.917.466,910419
> 9,540–14,31010.91102.1511.152,858273
> 14,310–23,85015.85118.2818.751,613224
> 23,85023.80166.1639.54990226

Monthly total CU income including asset variation; historical BRL. Weighted incidence among all CUs in band; conditional value among positive 90-day CUs. Values at 15 January 2018 prices. Monetary modes 1–6. Counts unweighted; point estimates only.

Point estimates only. Primary monetary acquisition modes 1–6; official processing includes imputation. Amounts at 15 January 2018 reference.

Download chart (SVG) ↓ · Download chart (PNG) ↓

Accounting for the highest–lowest income-band gap

Symmetric point decomposition on a shared ninety-day basis.

Full chart and source
Of R$39.08 gap, R$26.02 incidence component and R$13.06 conditional-value component; 66.59 percent incidence share.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

Symmetric highest-minus-lowest decomposition
Component90-day value (R$)Share of gap (%)
Incidence26.0266.59
Conditional value13.0633.41
Total mean gap39.08100.00

R$ per CU over 90 days at 15 January 2018 prices; monthly total CU income includes asset variation. Full-precision calculations precede rounding. Descriptive accounting, not causation.

Highest and lowest monthly total CU income bands including asset variation. Not causal or annual buyer incidence.

Download chart (SVG) ↓ · Download chart (PNG) ↓

Data & downloads

Scroll horizontally to see all columns.

All seven income bands: processed monetary non-textbook acquisition
Monthly CU income (R$)90-day incidence (%)Conditional 90-day value (R$)Mean 90-day value (R$)Sample CUsPositive CUs
≤ 1,9080.7759.860.4616,583130
> 1,908–2,8621.5656.670.8811,839154
> 2,862–5,7243.1394.922.9817,246470
> 5,724–9,5407.7895.917.466,910419
> 9,540–14,31010.91102.1511.152,858273
> 14,310–23,85015.85118.2818.751,613224
> 23,85023.80166.1639.54990226

Monthly total CU income including asset variation; historical BRL. Weighted incidence among all CUs in band; conditional value among positive 90-day CUs. Values at 15 January 2018 prices. Monetary modes 1–6. Counts unweighted; point estimates only.

Scroll horizontally to see all columns.

Symmetric highest-minus-lowest decomposition
Component90-day value (R$)Share of gap (%)
Incidence26.0266.59
Conditional value13.0633.41
Total mean gap39.08100.00

R$ per CU over 90 days at 15 January 2018 prices; monthly total CU income includes asset variation. Full-precision calculations precede rounding. Descriptive accounting, not causation.

Scroll horizontally to see all columns.

Own-consumer-unit monetary acquisition sensitivity: all income bands
Monthly CU income (R$)90-day incidence (%)Conditional 90-day value (R$)Mean 90-day value (R$)Positive CUs
≤ 1,9080.7560.060.45127
> 1,908–2,8621.4757.720.85147
> 2,862–5,7243.0695.802.93457
> 5,724–9,5407.6395.137.25410
> 9,540–14,31010.60103.1910.94265
> 14,310–23,85015.37120.1018.46215
> 23,85023.13168.5538.99221

Acquisition modes 1, 3 and 5 only; identical monthly total CU income bands including asset variation, price reference and ninety-day denominator. Not an observed-only sample.

Evidence & method

How we calculated it

  1. National consumer-unit frame from distinct resident keys; 58,039 CUs in 57,920 households. One PESO_FINAL per CU; retain units with no selected expense records.
  2. Prefixes 32017/32018 in block 32; primary monetary modes 1–6, own-unit sensitivity 1, 3, 5. Corrected V8000_DEFLA over 90 days; factor 4 only for explicitly annualized values.
  3. Seven official monthly total CU income bands including asset variation; corrected RENDA_TOTAL. All published revised weighted band counts reconcile after rounding.
  4. Weighted p=A/W, c=S/A, m=S/W; symmetric gap components (pH−pL)(cH+cL)/2 and (cH−cL)(pH+pL)/2. Independent point calculations agree within numerical tolerance.
  5. No main confidence intervals or significance claims. Approximate internal variance calculations omit FPC and extra imputation variance and do not establish official-equivalent inference.

What the data cannot tell us

  • Historical 2017–2018 only; amounts at 15 January 2018 prices, not present purchasing power. Income is not disposable earnings or wealth.
  • Whole-block and value imputation affect interpretation: 252 monetary records have a blank deflator; amount-flag exclusion does not create an observed-only sample.
  • Ninety-day positive acquisition is not annual buyer probability or reading participation. Monetary acquisition value differs from cash outflow and all books received.
  • No causal interpretation or claim that components represent intervention effects. Consumer-unit composition differs across income bands.
  • Sampling and nonsampling uncertainty remain; point agreement and published-count reconciliation do not eliminate recall or imputation error.

Original sources

  1. IBGE POF 2017–2018 public microdata

    IBGE POF 2017–2018 public microdata: Only DESPESA_INDIVIDUAL and MORADOR members used. Exact member CRC 32 and SHA 256 verified; no full-archive hash claimed. Public microdata, final calibrated weights, corrected income and acquisition values.

    Accessed October 30, 2026

  2. IBGE POF documentation, updated July 13, 2023

    IBGE POF documentation, updated July 13, 2023: Variable dictionary, product catalogue and collection manual. Block 32 pp.193–194: ninety-day non-textbook acquisition; monetary modes 1–6. Corrected amounts and blank-deflator block-imputation distinction.

    Accessed October 30, 2026

  3. IBGE POF table translators, updated July 13, 2023

    IBGE POF table translators, updated July 13, 2023: General category 110803 combines periodicals, books, magazines and coloring books. Narrow analysis retains only prefixes 32017 and 32018; translators use five-digit families.

    Accessed October 30, 2026

  4. IBGE POF 2017–2018: first results

    IBGE POF 2017–2018: first results: 2019 publication: income bands, sampling design, January 15, 2018 price reference, hot-deck imputation and annualization factor 4. Original band counts precede the documented income correction.

    Accessed October 30, 2026

  5. IBGE POF corrections and product information

    IBGE POF corrections and product information: Official indexed Erramos (errata notice): derived RENDA_TOTAL corrected October 29, 2019; income-class tables updated November 8, 2019, totals unchanged. Direct page fetch unavailable; indexed official content inspected.

    Accessed October 30, 2026

  6. IBGE Cultural Information and Indicators System 2007–2018

    IBGE Cultural Information and Indicators System 2007–2018: Table 28, p.110 official indexed text: all seven corrected consumer-unit population counts match after rounding. Direct PDF fetch unavailable; not represented as a downloaded source snapshot.

    Accessed October 30, 2026

  7. IBGE announcement of POF 2024–2025 first results

    IBGE announcement of POF 2024–2025 first results: Official indexed announcement schedules first release November 27, 2026, after source access date. This study deliberately concerns historical 2017–2018; verify newer release availability before later reuse.

    Accessed October 30, 2026

Cite & reuse this study

Alisson Bento. The income gap in Brazilian non-textbook acquisition: incidence and conditional value. Original secondary analysis of IBGE Pesquisa de Orçamentos Familiares 2017–2018, corrected public microdata. Sources accessed October 30, 2026. https://alissonbento.com/research/brazil-nontextbook-acquisition-income-decomposition-2017-2018/

This download set contains original national-band aggregate calculations, original code and explanatory methods by Alisson Bento, with source attribution. It does not redistribute raw microdata, source manuals, entire source reports, household/person/PSU identifiers, logos or confidential records. Users should cite IBGE POF 2017–2018 as the data source and Alisson Bento for this selection, decomposition and explanation. No official endorsement is implied. Consult the source's applicable terms before redistributing original files; this analysis does not grant additional rights in those files.

Corrections or editorial questions

Pass it on.

The income gap in Brazilian non-textbook acquisition: incidence and conditional value | Alisson Bentoalissonbento.com

WhatsAppFacebookXTelegramRedditThreadsEmailSMS