Group of Experts on Measuring Poverty and Inequality, UNECE, Geneva, 28-29 November 2022

Anna Bieńkuńska

Tomasz Piasecki

Comparison of determinants of subjective and objective poverty in Poland based on the results of the EU-SILC

2stat.gov.pl

1. Applied poverty measures 2. Distribution of poverty by selected characteristics

of households 3. Determinants of different kinds of poverty

(model based analysis) 4. Concluding remarks

Contents:

3stat.gov.pl

• subjective economic stress indicator (ECSTRESS) - indirect (proxy) subjective poverty measure

• at-risk-of-poverty-rate (ARPR) – objective relative measure of poverty based on income poverty threshold (60% of the national median equivalised disposable income after social transfers)

• severe material and social deprivation (SMSD) - objective absolute measure of non-monetary poverty (7 of 13 deprivation)

• severe material deprivation (SMD) - objective absolute measure of non-monetary poverty (4 of 9 deprivation)

Applied poverty measures

4stat.gov.pl

EU-SILC 2019 EU-SILC 2022

Households making ends meet with great difficulty

4.5 3.7

Households making ends meet with difficulty 11.2 11.7

Households making ends meet with some difficulty

34.7 39.8

Households making ends meet fairly easily 34.4 33.1

Households making ends meet easily 11.6 9.6

Households making ends meet very easily 3.6 2.0

ECSTRESS 15.7 15.4

Subjective assessments of ability ”to make ends meet” and subjective economic stress indicator (ECSTRESS), % of people

ECSTRESS - share of people living in households making ends meet with difficulty or great difficulty

5stat.gov.pl

13.0% 13.1%

7.1%

18.8%

37.8% 39.5%

6.6%

37.7%

10.7%

23.3%

42.3%

60.3%

1.9% 0.5% 0.4%

3.8%

9.6% 12.6%

0%

10%

20%

30%

40%

50%

60%

70%

Employees Farmers Self-employed Retirees Pensioners Living on unearned sources

Subjective economic stress indicator At-risk-of-poverty rate (ARPR) Severe material and social deprivation (SMSD)

Poverty indicators by socio-economic groups of households (main source of income), EU-SILC 2022

6stat.gov.pl

Poverty indicators by household type, EU-SILC 2022

20.5% 22.4% 22.5%

29.9%

12.9%

15.9% 15.6%

34.5%

10.5% 10.3%

17.1%

14.8% 16.4%

25.4%

20.4%

23.6%

35.6%

8.9% 10.5% 10.5%

22.2%

10.2% 11.8%

20.5%

12.9%

16.7%

8.1% 6.5% 6.3% 6.0%

3.1% 3.9% 2.3%

9.3%

1.9% 1.7% 2.5% 1.4%

7.1%

0%

5%

10%

15%

20%

25%

30%

35%

40%

Single male

(age<65)

Single female

(age<65)

Single male

(age 65+)

Single female

(age 65+)

2 adults, no child (age<65)

*

2 adults, no child

(age 65+) **

At least 3 adults, no child

1 adult, at least 1 child

2 adults, 1 child

2 adults, 2 children

2 adults, at least

3 children

At least 3 adults, at least 1 child

Other

Subjective economic stress indicator At-risk-of-poverty rate (ARPR) Severe material and social deprivation (SMSD)

* both persons aged less than 65 ** at least one person aged 65+

7stat.gov.pl

Poverty indicators by the type of locality, EU-SILC 2022

11.4%

13.7%

15.4%

18.8%

16.0%

5.9%

9.1% 10.2%

13.5%

19.2%

1.8%

3.4% 2.5%

3.8% 2.7%

0%

5%

10%

15%

20%

25%

500 thous. or more 100-500 thous. 20-100 thous. up to 20 thous. Rural areas

Subjective economic stress indicator At-risk-of-poverty rate (ARPR) Severe material and social deprivation (SMSD)

Urban areas by population

8stat.gov.pl

EU-SILC 2019 EU-SILC 2022

Subjective economic stress indicator (ECSTRESS) 15.7 15.4

At-risk-of-poverty-rate (ARPR) 15.4 13.7

Severe material and social deprivation (SMSD) 3.5 2.8

Severe material deprivation (SMD) 3.6 3.0

Comparison of poverty rates between EU-SILC 2019 and 2022

9stat.gov.pl

Comparison of poverty distribution by socio-economic groups of households between EU-SILC 2019 and 2022

12.7%

13.3%

7.9%

22.6%

44.1%

45.6%

13.0%

13.1%

7.1%

18.8%

37.8%

39.5%

0% 10% 20% 30% 40% 50%

Employees

Farmers

Self-employed

Retirees

Pensioners

Living on unearned

sources

2019 2022

7.7%

34.9%

16.5%

25.0%

50.3%

57.1%

6.6%

37.7%

10.7%

23.3%

42.3%

60.3%

0% 20% 40% 60%

2019 2022

2.5%

2.2%

1.4%

4.9%

17.4%

19.7%

2.2%

2.8%

0.8%

3.9%

10.8%

10.9%

0% 5% 10% 15% 20%

2019 2022

Subjective economic stress indicator Severe material deprivationARPR

10stat.gov.pl

Comparison of poverty distribution by household type between EU-SILC 2019 and 2022

Subjective economic stress indicator Severe material deprivationARPR

29.4%

25.6%

24.7%

34.4%

13.5%

16.7%

16.4%

35.0%

12.5%

9.8%

14.5%

13.2%

16.5%

20.5%

22.4%

22.5%

29.9%

12.9%

15.9%

15.6%

34.5%

10.5%

10.3%

17.1%

14.8%

16.4%

0% 10% 20% 30% 40%

Single male (age<65)

Single female (age<65)

Single male (age 65+)

Single female (age 65+)

2 adults, no child (age<65)

2 adults, no child (age 65+)

At least 3 adults, no child

1 adult, at least 1 child

2 adults, 1 child

2 adults, 2 children

2 adults, at least 3 children

At least 3 adults, at least 1 child

Other

2019 2022

35.9%

26.3%

26.7%

36.6%

13.4%

11.9%

11.4%

25.7%

10.8%

11.7%

13.8%

15.7%

20.0%

25.4%

20.4%

23.6%

35.6%

8.9%

10.5%

10.5%

22.2%

10.2%

11.8%

20.5%

12.9%

16.7%

0% 10% 20% 30% 40%

2019 2022

13.9%

8.1%

7.0%

7.8%

3.5%

3.9%

3.8%

9.7%

2.5%

1.4%

3.5%

2.3%

4.1%

8.1%

5.7%

5.8%

6.2%

3.7%

3.5%

3.1%

12.2%

1.7%

1.5%

1.6%

2.0%

4.3%

0% 5% 10% 15%

2019 2022

11stat.gov.pl

Comparison of poverty distribution by the type of locality between EU-SILC 2019 and 2022

Subjective economic stress indicator Severe material deprivationARPR

14.2%

16.7%

18.6%

18.1%

20.9%

11.4%

13.7%

15.4%

18.8%

16.0%

0% 5% 10% 15% 20% 25%

500 thous. or more

100-500 thous.

20-100 thous.

up to 20 thous.

Rural areas

2019 2022

11.3%

12.7%

14.3%

17.6%

26.4%

5.9%

9.1%

10.2%

13.5%

19.2%

0% 10% 20% 30%

2019 2022

3.7%

4.0%

4.5%

5.1%

5.4%

1.7%

3.1%

2.4%

4.4%

3.0%

0% 2% 4% 6%

2019 2022

Urban areas by population

12stat.gov.pl

• Logistic regression models • Households treated as statistical units (observations)

• Variables of interest: · subjective economic stress indicator (ECSTRESS, indirect subjective poverty measure) · at-risk-of-poverty-ratio (ARPR, relative income poverty objective measure) · severe material and social deprivation (SMSD, measure of needs deprivation,

objective absolute assessment of material situation)

• Explanatory variables (factors): · socio-economic group (main source of household income) · household type (household composition) · place of residence (type of locality, NUTS-2) · reference person (head of the household) characteristics (age, education level,

profession) · presence in the household of person with some individual socio-economic

characteristics: · unemployed person · person without independent (own) source of income (including children) · person with biological disability (taking into account the level of disability)

Determinants of different kinds of poverty Model based analysis

13stat.gov.pl

Logistic regression results (EU-SILC 2022) Aggregate influence of explanatory variables (factors)

Explanatory variable (factor) Degr. of

freedom

Variable of interest

(kind of poverty to be explained)

ECSTRESS ARPR SMSD

Wald stat. p-value Wald stat. p-value Wald stat. p-value

Socio-economic group 5 123.7 <0.001 878.1 <0.001 98.7 <0.001

Household type 9 466.9 <0.001 1167.8 <0.001 121.5 <0.001

Reference person characteristics:

age 5 17.5 0.004 130.8 <0.001 28.9 <0.001

education level 9 148.6 <0.001 123.9 <0.001 77.4 <0.001

profession 10 132.4 <0.001 411.1 <0.001 107.5 <0.001

Presence in the household of specific

person:

unemployed 1 151.9 <0.001 232.5 <0.001 137.2 <0.001

without independent source

of income 1 17.2 <0.001 405.1 <0.001 18.2 <0.001

with disability (not severe) 1 113.9 <0.001 0.7 0.408 72.8 <0.001

with severe disability 1 307.8 <0.001 4.5 0.034 151.9 <0.001

Characteristics of the residence place

type of locality 4 18.8 0.001 57.8 <0.001 5.6 0.235

NUTS-2 16 80.4 <0.001 52.1 <0.001 26.6 0.046

14stat.gov.pl

Regression – detailed effects/parameters (EU-SILC 2022) Socio-economic group (main income source)

Socio-economic group

Variable of interest

(kind of poverty to be explained)

ECSTRESS ARPR SMSD

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Employees (ref.)

Farmers -0.06 0.720 1.30 <0.001 -0.25 0.542

Self-employed -0.81 <0.001 0.66 <0.001 -0.80 0.021

Retirees 0.15 0.031 1.75 <0.001 0.10 0.407

Pensioners 0.35 <0.001 2.21 <0.001 0.48 0.002

Living on unearned sources 0.83 <0.001 2.71 <0.001 1.24 <0.001

15stat.gov.pl

Regression – detailed effects/parameters (EU-SILC 2022) Household type

Household type (composition)

Variable of interest

(kind of poverty to be explained)

ECSTRESS ARPR SMSD

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Single male (ref.)

Single female 0.31 <0.001 0.47 <0.001 -0.12 0.388

2 adults. no child -0.71 <0.001 -1.55 <0.001 -0.73 <0.001

At least 3 adults. no child -1.00 <0.001 -2.15 <0.001 -1.21 <0.001

1 adult. at least 1 child 0.17 0.270 -2.17 <0.001 -0.57 0.026

2 adults. 1 child -0.91 <0.001 -2.55 <0.001 -1.15 <0.001

2 adults. 2 children -1.08 <0.001 -2.48 <0.001 -1.60 <0.001

2 adults. at least 3 children -0.73 <0.001 -2.34 <0.001 -1.67 <0.001

At least 3 adults. at least 1 child -1.26 <0.001 -3.45 <0.001 -2.18 <0.001

Other -0.66 <0.001 -2.07 <0.001 -0.72 0.003

16stat.gov.pl

Regression – detailed effects/parameters (EU-SILC 2022) Reference person education level

Reference person

education level

Variable of interest

(kind of poverty to be explained)

ECSTRESS ARPR SMSD

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

At least primary (ref.)

Basic vocational -0.24 <0.001 -0.16 0.028 -0.39 <0.001

General secondary

(without matriculation) -0.43 0.001 -0.22 0.124 -0.79 0.001

General secondary

(with matriculation) -0.58 <0.001 -0.44 <0.001 -0.79 <0.001

Vocational secondary

(without matriculation) -0.43 <0.001 -0.34 0.002 -0.70 <0.001

Vocational secondary

(with matriculation) -0.65 <0.001 -0.54 <0.001 -0.78 <0.001

Post-secondary -0.46 <0.001 -0.54 <0.001 -1.02 <0.001

Tertiary (bachelor degree) -0.73 <0.001 -0.69 <0.001 -1.29 <0.001

Tertiary (master's degree or higher) -1.16 <0.001 -1.23 <0.001 -1.82 <0.001

17stat.gov.pl

Regression – detailed effects/parameters (EU-SILC 2022) Additional individual characteristics

Presence in the household

of a person with specific

individual characteristics:

Variable of interest

(kind of poverty to be explained)

ECSTRESS ARPR SMSD

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Unemployed 0.96 <0.001 1.35 <0.001 1.37 <0.001

Without independent

income source 0.38 <0.001 2.15 <0.001 0.60 <0.001

With biological disability

(not severe)* 0.50 <0.001 0.04 0.408 0.77 <0.001

With severe biological

disability 1.02 <0.001 0.15 0.034 1.26 <0.001

* and there is no person with severe biological disability

18stat.gov.pl

Regression – detailed effects/parameters (EU-SILC 2022) Type of locality

Type of locality

(place of residence)

Variable of interest

(kind of poverty to be explained)

ECSTRESS ARPR SMSD

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Parameter

estimate

Signif.

(p-value)

Urban area (by population):

500 thous. or more (ref.)

100-500 thous. 0.11 0.272 0.07 0.572 0.33 0.087

20-100 thous. 0.27 0.004 0.15 0.176 0.10 0.573

up to 20 thous. 0.32 0.001 0.33 0.005 0.30 0.109

Rural area 0.28 0.002 0.53 <0.001 0.20 0.255

19stat.gov.pl

Main determinants of the poverty (by kind)

Explanatory variable (factor)

Variable of interest

(kind of poverty to be explained)

ECSTRESS ARPR SMSD

Wald stat. p-value Wald stat. p-value Wald stat. p-value

Socio-economic group 123.7 <0.001 (2) 878.1 <0.001 98.7 <0.001

Household type (1) 466.9 <0.001 (1) 1167.8 <0.001 (3) 121.5 <0.001

HH head education level (4) 148.6 <0.001 123.9 <0.001 77.4 <0.001

HH head profession 132.4 <0.001 (3) 411.1 <0.001 (4) 107.5 <0.001

Presence of unemployed (3) 151.9 <0.001 232.5 <0.001 (2) 137.2 <0.001

Presence of person without

independent source of income 17.2 <0.001 (4) 405.1 <0.001 18.2 <0.001

Presence of person with severe

disability (2) 307.8 <0.001 4.5 0.034 (1) 151.9 <0.001

For each of 3 kinds of poverty, four determinants with the highest contribution to the explanation are marked blue (the 1st the is the darkest). All the marked effects are significant at the level 0.001. It occurs also in case of the effects on another kind of poverty (except of the effect of person with severe disability on ARPR, which is, however, significant at the level 0.05)

20stat.gov.pl

The factors (among those included in the models) with the highest contribution to the explanation of the particular kinds of poverty (measured by Wald statistics) are:

• For subjective economic stress indicator (ECSTRESS) : household type, presence of a person with severe disability, presence of an unemployed person and level of education of the head of household

• For the at-risk-of-poverty rate (ARPR): type of household, socio-economic group of the household, occupation of head of household, presence of a person with severe disability

• For severe material and social deprivation (SMSD): presence of a person with severe disability, presence of an unemployed person, household type, household head occupation

The sets of main determinants of the analysed kinds of poverty are similar but not the same; there are also differences in the order of main determinants (by their importance).

Household type (by composition) is the only factor which is among the four most important determinants of each kind of poverty; it is the main determinant of ECSTRESS and ARPR and the third in case of SMSD.

All explanatory factors included in the model have significant effects (at the level 0.05 at least) with except for evidently weaker relationships in case of the effect of disability on ARPR (only severe disability is significant) and the effect of residence place on SMSD (type of locality turned out to be insignificant – what means that the other factors fully explain the differences observed in the distribution)

Concluding remarks – main determinants of particular kinds of poverty

stat.gov.pl

Anna Bieńkuńska

[email protected]

Tomasz Piasecki

[email protected]

Thank you for your attention

Группа экспертов по измерению бедности и неравенства, ЕЭК ООН, Женева, 28-29 ноября 2022 г.

Анна Бенькуньска

Томаш Пясецки

Сравнение детерминант субъективной и объективной бедности в Польше на основе результатов исследования EU-SILC

2stat.gov.pl

1. Прикладные показатели бедности 2. Распределение бедности по отдельным характеристикам домохозяйств 3. Детерминанты различных видов бедности (анализ на основе моделей) 4. Заключительные замечания

Содержание:

3stat.gov.pl

• индикатор субъективного экономического стресса (ECSTRESS) - косвенный (прокси) субъективный показатель бедности

• уровень риска бедности (ARPR) - объективный относительный показатель бедности, основанный на пороге бедности по доходу (60% национального медианного эквивалентного располагаемого дохода после социальных выплат)

• тяжелая материальная и социальная депривация (СМСР) - объективный абсолютный показатель немонетарной бедности (7 из 13 деприваций)

• тяжелая материальная депривация (ТМД) - объективный абсолютный показатель немонетарной бедности (4 из 9 деприваций)

Прикладные показатели бедности

4stat.gov.pl

EU-SILC 2019 EU-SILC 2022

Домохозяйства, сводящие концы с концами с большим трудом

4.5 3.7

Домохозяйства, с трудом сводящие концы с концами

11.2 11.7

Домохозяйства, с некоторым трудом сводящие концы с концами

34.7 39.8

Домохозяйства, сравнительно легко сводящие концы с концами

34.4 33.1

Домохозяйствам легко сводящие концы с концами

11.6 9.6

Домохозяйства, легко сводящие концы с концами

3.6 2.0

ECSTRESS 15.7 15.4

Субъективные оценки способности «сводить концы с концами» и индикатор субъективного экономического стресса (ECSTRESS), % чел

ЭКСТРЕСС - доля людей, живущих в домохозяйствах, сводящих концы с концами с трудом или большими трудностями

5stat.gov.pl

13.0% 13.1%

7.1%

18.8%

37.8% 39.5%

6.6%

37.7%

10.7%

23.3%

42.3%

60.3%

1.9% 0.5% 0.4%

3.8%

9.6% 12.6%

0%

10%

20%

30%

40%

50%

60%

70%

Employees Farmers Self-employed Retirees Pensioners Living on unearned sources

Subjective economic stress indicator At-risk-of-poverty rate (ARPR) Severe material and social deprivation (SMSD)

Показатели бедности по социально-экономическим группам домохозяйств (основной источник дохода), EU- SILC 2022

6stat.gov.pl

Показатели бедности по типам домохозяйств, EU-SILC 2022

20.5% 22.4% 22.5%

29.9%

12.9%

15.9% 15.6%

34.5%

10.5% 10.3%

17.1%

14.8% 16.4%

25.4%

20.4%

23.6%

35.6%

8.9% 10.5% 10.5%

22.2%

10.2% 11.8%

20.5%

12.9%

16.7%

8.1% 6.5% 6.3% 6.0%

3.1% 3.9% 2.3%

9.3%

1.9% 1.7% 2.5% 1.4%

7.1%

0%

5%

10%

15%

20%

25%

30%

35%

40%

Single male

(age<65)

Single female

(age<65)

Single male

(age 65+)

Single female

(age 65+)

2 adults, no child (age<65)

*

2 adults, no child

(age 65+) **

At least 3 adults, no child

1 adult, at least 1 child

2 adults, 1 child

2 adults, 2 children

2 adults, at least

3 children

At least 3 adults, at least 1 child

Other

Subjective economic stress indicator At-risk-of-poverty rate (ARPR) Severe material and social deprivation (SMSD)

* оба человека в возрасте до 65 лет ** как минимум один человек в возрасте 65+

7stat.gov.pl

Показатели бедности по типу населенного пункта, EU-SILC 2022

11.4%

13.7%

15.4%

18.8%

16.0%

5.9%

9.1% 10.2%

13.5%

19.2%

1.8%

3.4% 2.5%

3.8% 2.7%

0%

5%

10%

15%

20%

25%

500 thous. or more 100-500 thous. 20-100 thous. up to 20 thous. Rural areas

Subjective economic stress indicator At-risk-of-poverty rate (ARPR) Severe material and social deprivation (SMSD)

Городские районы по численности населения

8stat.gov.pl

EU-SILC 2019 EU-SILC 2022

Индикатор субъективного экономического стресса (ECSTRESS) 15.7 15.4

Уровень риска бедности (ARPR) 15.4 13.7

Тяжелые материальные и социальные лишения (SMSD) 3.5 2.8

Тяжелые материальные лишения (SMD) 3.6 3.0

Сравнение уровня бедности между EU-SILC 2019 и 2022 гг.

9stat.gov.pl

Сравнение распределения бедности по социально- экономическим группам домохозяйств между EU-SILC 2019 и 2022 гг.

12.7%

13.3%

7.9%

22.6%

44.1%

45.6%

13.0%

13.1%

7.1%

18.8%

37.8%

39.5%

0% 10% 20% 30% 40% 50%

Employees

Farmers

Self-employed

Retirees

Pensioners

Living on unearned

sources

2019 2022

7.7%

34.9%

16.5%

25.0%

50.3%

57.1%

6.6%

37.7%

10.7%

23.3%

42.3%

60.3%

0% 20% 40% 60%

2019 2022

2.5%

2.2%

1.4%

4.9%

17.4%

19.7%

2.2%

2.8%

0.8%

3.9%

10.8%

10.9%

0% 5% 10% 15% 20%

2019 2022

Субъективный индикатор экономического стресса Тяжелые материальные лишенияARPR

10stat.gov.pl

Сравнение распределения бедности по типам домохозяйств между ЕС-СИЛК в 2019 и 2022 годах

Субъективный индикатор экономического стресса Тяжелые материальные лишенияARPR

29.4%

25.6%

24.7%

34.4%

13.5%

16.7%

16.4%

35.0%

12.5%

9.8%

14.5%

13.2%

16.5%

20.5%

22.4%

22.5%

29.9%

12.9%

15.9%

15.6%

34.5%

10.5%

10.3%

17.1%

14.8%

16.4%

0% 10% 20% 30% 40%

Single male (age<65)

Single female (age<65)

Single male (age 65+)

Single female (age 65+)

2 adults, no child (age<65)

2 adults, no child (age 65+)

At least 3 adults, no child

1 adult, at least 1 child

2 adults, 1 child

2 adults, 2 children

2 adults, at least 3 children

At least 3 adults, at least 1 child

Other

2019 2022

35.9%

26.3%

26.7%

36.6%

13.4%

11.9%

11.4%

25.7%

10.8%

11.7%

13.8%

15.7%

20.0%

25.4%

20.4%

23.6%

35.6%

8.9%

10.5%

10.5%

22.2%

10.2%

11.8%

20.5%

12.9%

16.7%

0% 10% 20% 30% 40%

2019 2022

13.9%

8.1%

7.0%

7.8%

3.5%

3.9%

3.8%

9.7%

2.5%

1.4%

3.5%

2.3%

4.1%

8.1%

5.7%

5.8%

6.2%

3.7%

3.5%

3.1%

12.2%

1.7%

1.5%

1.6%

2.0%

4.3%

0% 5% 10% 15%

2019 2022

11stat.gov.pl

Сравнение распределения бедности по типам населенных пунктов между ЕС-СИЛК 2019 и 2022 гг.

Субъективный индикатор экономического стресса Тяжелые материальные лишенияARPR

14.2%

16.7%

18.6%

18.1%

20.9%

11.4%

13.7%

15.4%

18.8%

16.0%

0% 5% 10% 15% 20% 25%

500 thous. or more

100-500 thous.

20-100 thous.

up to 20 thous.

Rural areas

2019 2022

11.3%

12.7%

14.3%

17.6%

26.4%

5.9%

9.1%

10.2%

13.5%

19.2%

0% 10% 20% 30%

2019 2022

3.7%

4.0%

4.5%

5.1%

5.4%

1.7%

3.1%

2.4%

4.4%

3.0%

0% 2% 4% 6%

2019 2022

Urban areas by population

12stat.gov.pl

• Модели логистической регрессии • Домохозяйства, рассматриваемые как статистические единицы (наблюдения)

• Переменные, представляющие интерес : · показатель субъективного экономического стресса (ECSTRESS, косвенная

субъективная мера бедности) · коэффициент риска бедности (ARPR, объективная мера бедности по

относительному доходу) · тяжелые материальные и социальные лишения (SMSD, мера лишения

потребностей, объективная абсолютная оценка материального положения)

• Объясняющие переменные (факторы): · социально-экономическая группа (основной источник дохода домохозяйства) · тип домохозяйства (состав домохозяйства) · место жительства (тип населенного пункта, NUTS-2) · характеристики референта (главы домохозяйства) (возраст, уровень

образования, профессия) · наличие в домохозяйстве человека с определенными индивидуальными

социально-экономическими характеристиками : · безработный · лицо, не имеющее независимого (собственного) источника дохода (включая детей) · человек с биологической инвалидностью (с учетом уровня инвалидности)

Детерминанты различных видов бедности Анализ на основе моделей

13stat.gov.pl

Результаты логистической регрессии (EU-SILC 2022) Совокупное влияние объясняющих переменных (факторов)

Объясняющая переменная

(фактор)

Степень

свободы

Переменная интереса

(вид бедности, который необходимо объяснить)

ECSTRESS ARPR SMSD

Wald stat. p-value Wald stat. p-value Wald stat. p-value

Социоэкономическая группа 5 123.7 <0.001 878.1 <0.001 98.7 <0.001

Тип домохозяйства 9 466.9 <0.001 1167.8 <0.001 121.5 <0.001

Характеристики референтного лица:

возраст 5 17.5 0.004 130.8 <0.001 28.9 <0.001

уровень образования 9 148.6 <0.001 123.9 <0.001 77.4 <0.001

профессия 10 132.4 <0.001 411.1 <0.001 107.5 <0.001

Присутствие в семье конкретного

человека:

безработный 1 151.9 <0.001 232.5 <0.001 137.2 <0.001

без независимого

источника дохода 1 17.2 <0.001 405.1 <0.001 18.2 <0.001

с инвалидностью (не

тяжелой) 1 113.9 <0.001 0.7 0.408 72.8 <0.001

с тяжелой формой

инвалидности 1 307.8 <0.001 4.5 0.034 151.9 <0.001

Характеристика места проживания

тип населенного пункта 4 18.8 0.001 57.8 <0.001 5.6 0.235

NUTS-2 16 80.4 <0.001 52.1 <0.001 26.6 0.046

14stat.gov.pl

Регрессия - подробные эффекты/параметры (EU-SILC 2022) Социоэкономическая группа (основной источник дохода)

Социоэкономическая

группа

Переменная интереса

(вид бедности, который необходимо объяснить)

ECSTRESS ARPR SMSD

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимост

ь (p-value)

Сотрудники (реф.)

Фермеры -0.06 0.720 1.30 <0.001 -0.25 0.542

Самозанятые -0.81 <0.001 0.66 <0.001 -0.80 0.021

Отставники 0.15 0.031 1.75 <0.001 0.10 0.407

Пенсионеры 0.35 <0.001 2.21 <0.001 0.48 0.002

Жизнь на нетрудовые

доходы 0.83 <0.001 2.71 <0.001 1.24 <0.001

15stat.gov.pl

Регрессия - подробные эффекты/параметры (EU-SILC 2022) Тип домохозяйства

Тип домохозяйства (состав)

Переменная интереса

(вид бедности, который необходимо объяснить)

ECSTRESS ARPR SMSD

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимост

ь (p-value)

Оценка

параметра

Значимост

ь (p-value)

Одинокий мужчина (ссылка)

Одинокая женщина 0.31 <0.001 0.47 <0.001 -0.12 0.388

2 взрослых без детей -0.71 <0.001 -1.55 <0.001 -0.73 <0.001

Не менее 3 взрослых без детей -1.00 <0.001 -2.15 <0.001 -1.21 <0.001

1 взрослый, минимум 1 ребенок 0.17 0.270 -2.17 <0.001 -0.57 0.026

2 взрослых, 1 ребёнок -0.91 <0.001 -2.55 <0.001 -1.15 <0.001

2 взрослых, 2 детей -1.08 <0.001 -2.48 <0.001 -1.60 <0.001

2 взрослых, минимум 3 ребенка -0.73 <0.001 -2.34 <0.001 -1.67 <0.001

Не менее 3 взрослых, минимум

1 ребенок -1.26 <0.001 -3.45 <0.001 -2.18 <0.001

Прочее -0.66 <0.001 -2.07 <0.001 -0.72 0.003

16stat.gov.pl

Регрессия - подробные эффекты/параметры (EU-SILC 2022) Уровень образования референта

Уровень образования

референта

Переменная интереса

(вид бедности, который необходимо объяснить)

ECSTRESS ARPR SMSD

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимо

сть (p-

value)

По меньшей мере, начальное

(ссылка)

Базовое профессиональное -0.24 <0.001 -0.16 0.028 -0.39 <0.001

Общее среднее образование (без

аттестата зрелости) -0.43 0.001 -0.22 0.124 -0.79 0.001

Общее среднее (с аттестатом

зрелости) -0.58 <0.001 -0.44 <0.001 -0.79 <0.001

Среднее профессиональное

образование (без аттестата

зрелости)

-0.43 <0.001 -0.34 0.002 -0.70 <0.001

Среднее специальное образование

(с получением аттестата

зрелости)

-0.65 <0.001 -0.54 <0.001 -0.78 <0.001

Высшее образование -0.46 <0.001 -0.54 <0.001 -1.02 <0.001

Высшее образование (бакалавриат) -0.73 <0.001 -0.69 <0.001 -1.29 <0.001

Высшее образование (степень

магистра или выше) -1.16 <0.001 -1.23 <0.001 -1.82 <0.001

17stat.gov.pl

Регрессия - подробные эффекты/параметры (EU-SILC 2022) Дополнительные индивидуальные характеристики

Наличие в домохозяйстве

человека с

определенными

индивидуальными

характеристиками:

Переменная интереса

(вид бедности, который необходимо объяснить)

ECSTRESS ARPR SMSD

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимос

ть (p-

value)

Безработные 0.96 <0.001 1.35 <0.001 1.37 <0.001

Без независимого

источника дохода 0.38 <0.001 2.15 <0.001 0.60 <0.001

С биологической

инвалидностью (не

тяжелой)*

0.50 <0.001 0.04 0.408 0.77 <0.001

С тяжелой биологической

инвалидностью 1.02 <0.001 0.15 0.034 1.26 <0.001

* и нет человека с тяжелой биологической инвалидностью

18stat.gov.pl

Регрессия - подробные эффекты/параметры(EU-SILC 2022) Тип населенного пункта

Тип населенного пункта

(место жительства)

Переменная интереса

(вид бедности, который необходимо объяснить)

ECSTRESS ARPR SMSD

Оценка

параметра

Значимост

ь (p-value)

Оценка

параметра

Значимос

ть (p-

value)

Оценка

параметра

Значимос

ть (p-

value)

Городская зона (по

численности населения):

500 тыс. и более (реф.)

100-500 тыс. 0.11 0.272 0.07 0.572 0.33 0.087

20-100 тыс. 0.27 0.004 0.15 0.176 0.10 0.573

up to 20 тыс. 0.32 0.001 0.33 0.005 0.30 0.109

Сельская зона 0.28 0.002 0.53 <0.001 0.20 0.255

19stat.gov.pl

Основные факторы, определяющие бедность (по видам)

Объяснительная переменная

(фактор)

Переменная интереса

(вид бедности, который необходимо объяснить)

ECSTRESS ARPR SMSD

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимость

(p-value)

Оценка

параметра

Значимос

ть (p-

value)

Социоэкономическая группа 123.7 <0.001 (2) 878.1 <0.001 98.7 <0.001

Тип домохозяйства (1) 466.9 <0.001 (1) 1167.8 <0.001 (3) 121.5 <0.001

Уровень образования главы

домохозяйства (4) 148.6 <0.001 123.9 <0.001 77.4 <0.001

Профессия главы

домохозяйства 132.4 <0.001 (3) 411.1 <0.001 (4) 107.5 <0.001

Наличие безработных (3) 151.9 <0.001 232.5 <0.001 (2) 137.2 <0.001

Наличие лица без

независимого источника

дохода 17.2 <0.001 (4) 405.1 <0.001 18.2 <0.001

Наличие лица с тяжелой

инвалидностью (2) 307.8 <0.001 4.5 0.034 (1) 151.9 <0.001

Для каждого из трех видов бедности четыре детерминанты с наибольшим вкладом в объяснение отмечены синим цветом (первая - самая темная). Все отмеченные эффекты значимы на уровне 0,001. Это происходит и в случае эффектов на другой вид бедности (за исключением эффекта человека с тяжелой инвалидностью на ARPR, который, однако, значим на уровне 0,05)

20stat.gov.pl

Факторы (из числа включенных в модели) с наибольшим вкладом в объяснение конкретных видов бедности (по статистике Вальда) следующие : • Для субъективного показателя экономического стресса (ECSTRESS): тип домохозяйства,

наличие человека с тяжелой инвалидностью, наличие безработного и уровень образования главы домохозяйств

• Для коэффициента риска бедности (ARPR): тип домохозяйства, социально-экономическая группа домохозяйства, профессия главы домохозяйства, наличие человека с тяжелой инвалидностью

• Для тяжелых материальных и социальных лишений (ТМСЛ): наличие людей с тяжелой инвалидностью, наличие безработных, тип домохозяйства, род занятий главы домохозяйства

Наборы основных детерминант анализируемых видов бедности схожи, но не одинаковы; существуют также различия в порядке расположения основных детерминант (по их значимости).

Тип домохозяйства (по составу) - единственный фактор, который входит в четверку наиболее важных детерминант каждого вида бедности; он является основным детерминантом для ECSTRESS и ARPR и третьим в случае SMSD.

Все объясняющие факторы, включенные в модель, имеют значимые эффекты (на уровне не ниже 0,05), за исключением явно слабых связей в случае влияния инвалидности на ARPR (значима только тяжелая инвалидность) и влияния места проживания на SMSD (тип населенного пункта оказался незначимым - что означает, что остальные факторы полностью объясняют различия, наблюдаемые в распределении)

Заключительные замечания - основные детерминанты отдельных видов бедности

stat.gov.pl

Anna Bieńkuńska

[email protected]

Tomasz Piasecki

[email protected]

Благодарим за внимание

United Nations Economic Commission for Europe

Palais des Nations, 1211 Geneva 10, Switzerland

Telephone: +41 (0)22 917 1234 (Palais des Nations)

UNITED NATIONS

ECONOMIC COMMISSION FOR EUROPE

CONFERENCE OF EUROPEAN STATISTICIANS

Group of Experts on Measuring Poverty and Inequality

28-29 November 2023

Workshop on Harmonization of Poverty Statistics to

Measure SDG 1 and 10

27 November 2023

Title of contribution Comparison of determinants of subjective and objective poverty in Poland

based on the results of the EU-SILC

Author Name(s) Anna Bienkunska, Tomasz Piasecki

Presenter Name Anna BIenkunska

Presenter Organization Statistics Poland

Topic Subjective poverty

Summary:

EU-SILC, the European survey on income and living conditions of the population, is a basic source of

information on poverty that is harmonized at European Union level. The presentation will demonstrate

the results of an analysis based on this survey that compares the determinants of subjective and

objective poverty in Poland. In the EU-SILC, there are no questions that directly measure subjective

poverty (the share of the population that considers itself poor). Instead, the variables included in this

survey provide an opportunity to use indirect (proxy) methods to measure subjective poverty. On the

basis of a question on the assessment of the ability to make ends meet (variable HS 120), an indicator

of subjective economic stress was calculated, which is treated as an indirect measure of subjective

poverty. In the case of so-called objective indicators, the analysis includes two indicators - „at-risk-of-

poverty rate” (ARPR) and the 'severe material and social deprivation rate' (SMSD). Logistic regression

models were used to analyze the impact of potential factors on the occurrence of various forms of

poverty. EU-SILC 2022 (latest available data) and EU-SILC 2019 (the period before COVID-19) were

used in the poverty determinant analysis. The repeated measurement (and analysis) referring to two

time points can, to some extent, assess the stability of poverty risk factors.

Please select your preferred contribution (you may select both options):

☒ Presentation

☐ Paper (to be submitted by 20 October)

1

FUNITED NATIONS ECONOMIC COMMISSION FOR EUROPE

CONFERENCE OF EUROPEAN STATISTICIANS

Expert Meeting on Statistical Data Confidentiality

26-28 September 2023, Wiesbaden

SDC in statistical education - the Polish experience

Tomasz Klimanek1,2, Tomasz Józefowski1,2, Andrzej Młodak1,3 1Statistical Office in Poznań, Poland, 2Poznań University of Economics and Business, Poznań, Poland, 3Calisia University – Kalisz, Poland

Abstract

In addition to the development of the principles and tools of statistical disclosure control, it is also important to raise public awareness of why and how these methods are applied. In particular, the public should be educated about the use of SDC to protect privacy on the one hand and to maximize the amount of publicly available information, on the other. Educational efforts in this regard should obviously start with the NSI staff responsible for efficient data protection, and then this knowledge should be disseminated among data users and taught to people who handle statistics and data managers in various institutions and economic entities. In the paper we present some activities in this area that have been undertaken in Poland recently. They include specialized training workshops for members of the team responsible for SDC methods, including current and future experts in this field at Statistics Poland as well as SDC topics covered in the preparatory training for new employees of statistical offices in Poland. We also offer a brief overview of a monograph on SDC methods, which is about to be published by Poznań University of Economics and Business (PUEB), which is addressed to all those interested in ways of handling sensitive data or applying for access to microdata or tabular data from various sources. We will also present the syllabus of a new course entitled “Data confidentiality protection methods”, which is going to be launched at PUEB. Other educational initiatives, such as a tutorial on data protection for students writing their diploma theses are mentioned.

Key words: statistical disclosure control, statistical education, training, preparatory training

1. Introduction

The growing demand for detailed statistical information in recent years has led to the intensive development of statistical disclosure control (SDC) methods, processes, and tools. The main purpose of SDC is to ensure safe and efficient dissemination of data collected by official statistics. It is therefore necessary to prepare employees of national statistical institutes to efficiently implement SDC processes. NSI staff should have a broad knowledge of various SDC methods and tools, be able to apply them in specific situations and obtain sufficiently safe and useful data to be released. In particular, they should have the ability to choose appropriate SDC methods to handle data from a given survey taking into account its methodology and current (and sometimes even expected future) users’ needs.

Also newly employed staff could benefit from SDC training. In Poland – as in many other countries – new employees are obliged to participate in a special training, which is usually the first stage of their professional career in public administration and civil service. We believe this training should cover problems connected with statistical confidentiality and SDC, even if only to a limited extent.

Official statistics are not the only data source that can be used for scientific or analytical purposes. There are many data custodians (both public, such as central government agencies and units of local government, and private, e.g. various economic entities), which collect data they need to perform their activities and tasks. For instance, data from public opinion polls or market research concerning preferences of customers (e.g. audience size of different TV stations or the structure of purchases based on data from cash registers) are very valuable for social studies and are of interest to scientists and analysts. This means that the knowledge of efficient ways of protecting the privacy of their stakeholders or customers is also important for them. In addition, the protection of

2

data privacy is regulated by various laws, which also require commonly available guidelines. To exemplify how such guidelines could look like, we present a Polish handbook written by current or former NSI employees, which we believe could provide useful knowledge about SDC to various data holders planning to release their data to external users.

The last but not least important group of persons who should be aware of the benefits, drawbacks and expected effects of SDC methods are end users of statistical data. Many of them are students preparing their diploma theses (mainly BSc., MSc or PhD), scientists conducting their research or various business analysts (e.g. in the field of market research). When users are familiar with the principles regulating access to data, general types of SDC methods (e.g. confidentiality rules) and the impact of SDC on data quality, they can then correctly assess the real value of resulting data analyses. Of course, it is difficult to reach all potential recipients with the right message, but by providing relevant information to students one can hope that in time this knowledge will reach all interested parties. Moreover, students often conduct their own polls or targeted surveys and should know how to properly protect sensitive information they collect. This is why we also present a proposal of a course syllabus called “Methods of protecting data privacy” addressed to students of economic studies who cover quantitative methods at the Poznań University of Economics and Business in Poznań, Poland.

The paper is organized as follows. In Section we present the aims and scope of trainings for experts, which cover the main principles of statistical disclosure control and the main stages of the SDC process. Section 3 describes how basic aspects of SDC are presented during the initial training of new NSI employees. Section 4 deals with selected problems and examples of SD education for a wide range of potential users. Section 5 provides the most important conclusions od suggestions for the future.

2. Training of experts

SDC education should obviously start with the NSI staff responsible for efficient data protection. The goal is to provide the staff with a broad knowledge of SDC and how these methods apply to survey methodology in order to maximally protect the privacy of respondents while simultaneously ensuring the maximum utility of data for end users. Basic knowledge about SDC should also be disseminated among data users and taught to people who handle statistics, e.g. data managers in various institutions and economic entities.

The way the SDC process is organised varies across countries. In Poland, there is a dedicated team responsible for SDC methods, which prepares the principles and guidelines for how to conduct the SDC process in various surveys. Moreover, units responsible for particular surveys should have their own methodologists that are responsible for SDC. To ensure that all NSI units in Poland have the required knowledge of SDC, a series of trainings have been conducted. They mainly took place in 2019 and were divided into basic and advanced courses. The basic training covered the following topics:

 general concepts and their definitions: o types of data disclosure o types of users from the point of view of SDC

 regulations and principles used in the international practice of SDC: o regulations in Poland, the European Union and other countries around the world o codes of good practices

 basic types of disclosed information – tabular data and microdata: o types of disclosed information o types of microdata from the point of view of SDC

 disclosure risk and information loss: o disclosure scenarios o individual, global and hierarchical risk o measurement of risk o information loss due to the application of SDC o ways of measuring information loss o the trade-off between minimizing disclosure risk and minimizing information loss

 main methods of protecting sensitive information in frequency and magnitude tables: o frequency tables:

 non-perturbative methods  perturbative methods

3

o magnitude tables:  non-perturbative methods  perturbative methods

 SDC methods and techniques for microdata: o anonymization as a preliminary stage of protecting sensitive data o non-perturbative methods:

 subsampling  recoding  local data suppression

o perturbative methods:  noise addition  microaggregation  data or range swapping  rounding  secondary subsampling  PRAM

 software used for protecting statistical confidentiality: o τ–Argus o μ–Argus o tools of the R environment

 problems connected with the preparation of microdata for external users (public use files) – e.g. data from the LFS or EU-SILC:

o principles of data disclosure to specific users and their rights and obligations o organization and safety of microdata disclosure

 stationary access (in research data centres)  remote access  control of microdata use.

The advanced training included specific methodological and technical problems of SDC and relevant practical exercises, i.e.:

 application of dedicated R packages – e.g. sdcTable and sdcMicro,  methods of generating synthetic microdata and new algorithms for protecting statistical

confidentiality, e.g. the cell-key method,  problems connected with the preparation of microdata for external users (PUFs), e.g. from the LFS

or the EU-SILC,  methods of protecting statistical confidentiality of census data.

The third training was addressed to members of the general expert team. The topics covered were similar to those in the first two trainings, but they were presented in a more condensed manner. In a questionnaire after completing the training the participants said they were planning to use the presented methods and tools in their daily work. When asked to identify the main problems related to SDC, the mentioned things like the need for more training in the use of R, the lack of competent staff and the lack of experience in the use of SDC.

In the following years (among other things because of the COVID-19 pandemic) SDC training was offered in the form of direct individual consultations or on-line sessions, which also proved to be effective. Currently, the Team for Methods of Statistical Disclosure Control is preparing special guidelines for NSI employees (Statistics Poland (2023)) to support the implementation of the SDC process.

3. SDC issues for new employees of official statistics

The development of competent NSI staff involves equipping new employees with the knowledge and skills they need to perform tasks in the course of civil service. Preparatory service training for people employed in public administration includes learning about basic mechanisms and rules that govern specific public agencies. In the case of NSI employees preparatory service training consists of two parts: obligatory and optional. The obligatory part is divided into basic and extended. The basic part is in fact common for all government institutions and concerns the general functioning of the public administration sector. The extended part covers specific aspects of

4

official statistics (the legal basis, organization of official statistics, organization of statistical surveys, IT, safety of information, crisis management, international cooperation, publication policy, etc.).

Since 2021, the methodological aspects of SDC have been covered as part of topic 3 (principles of organising statistical surveys): Confidentiality and utility of output data in official statistics1. The study materials succinctly describe various aspects of data protection and SDC. Given the scope of the preparatory service training SDC issues had to be limited to the most important ones and included:

 the legal basis of statistical confidentiality  anonymisation and pseudonymisation  aims and basic concepts of SDC  disclosure, disclosure risk and information loss  user as an analyst and user as an intruder  the SDC process, its components and stages  SDC for microdata  SDC for tabular data  SDC for statistical outputs.

The materials do not include any mathematical and formal descriptions and do not cover specific details of different SDC tools. The effectiveness of the training is evaluated by means of a quiz at the end of this section to check if the knowledge has been properly acquired. Long-term effects of this training can be only be assessed after a few years at the earliest.

4. Education of potential data users and other stakeholders

As stated at the beginning, tom make sure that SDC methods and tools are well understood and used effectively, the knowledge about SDC needs to be broadly popularised not only among the NSI staff but also among various data holders responsible for the protection of data confidentiality and among end users interested in conducting analyses based on statistical data. The knowledge of SDC is also essential for people who collect any kind of sensitive data as part of their own research.

In an effort to make such knowledge easily accessible, a group of NSI employees from the Statistical Office in Poznań have prepared a handbook entitled Confidentiality vs. utility of statistical information. Dilemmas of statistical disclosure control (Młodak et al., 2023). The monograph is an attempt to provide a comprehensive description of all aspects of SDC, including the goal and definition of statistical disclosure control, its formal and legal principles, particularly current regulations and international recommendations regarding data confidentiality, types of released data (including metadata, i.e., information about definitions of concepts that characterise units, reference periods of data collection and units for which data are collected, measurement methods, possible exceptions to rules of determining variables or missing data, explanations of known causes of deviations or gaps; paradata, i.e., information about the process by which the data were collected, which can be used to explain interpretations of the resulting characteristics of survey units as well as data not included in the survey itself but automatically collected during survey administration, such as the number of times a given respondent has visited the webpage with the survey form, the number of data changes made, the time taken to complete the form, etc. as well as other data). The disclosure process, types of data users from the perspective of SDC, typologies of statistical outputs with respect to the protection of sensitive information and the trade-off between disclosure risk and data utility are explained. One section of the monograph contains an overview of the most important SDC solutions and regulations implemented in different European countries and another one provides a description of the main kinds of microdata and the role of metadata and paradata in the SDC process.

In particular, the monograph describes SDC methods for microdata, tabular data and output checking, presenting characteristics of each type of data, their sources and fundamental principles of their protection, mathematical methods and IT tools used for this purpose as well as organisational and technological aspects of releasing statistical data, which are associated with the risk of unit re-identification or disclosure of sensitive information. In addition, it also describes measures of protecting confidentiality which can be applied to outputs of statistical analyses, such as descriptive statistics, estimates of econometric models or charts. The authors have attempted to present the complexity of SDC and various rules for estimating re-identification risk. A

1 The topic “safety of information” concerns only the legal basis of personal data protection and the system of safety management, which is not directly related to the methodology of SDC.

5

comprehensive assessment of the effects of applying SDC should include an estimate of disclosure risk and the expected information loss due to the suppression or perturbation of sensitive information.

The main purpose of SDC is to achieve an optimal trade-off between these two minimisation goals. Therefore, the authors of the monograph pay a lot of attention to the problems of measuring disclosure risk and information loss. Taking into account various disclosure scenarios, two main types of disclosure risk are discussed:

 internal risk – when a unit can potentially be re-identified only on the basis of data made available to the user;

 external risk – when the user has access to other sources of data, which can be linked with the released dataset; this risk is much harder to measure because there is usually little or no information about other data sources available to the user, except for what can be inferred from the user’s place of employment (e.g., if the user works at a labour office, they are likely to have access to the register of unemployed people, which can be linked with microdata from the LFS survey).

The monograph provides numerous examples showing how to construct such measures and how to interpret them; the examples demonstrate the usefulness of various measures for different use cases of released data, including measures of estimation precision.

Nowadays, data collection, processing and analysis cannot be performed without the help of appropriate IT tools. This is also true in the case of statistical disclosure control. Although the development of this branch of statistics is only now starting to accelerate, a number of useful programming tools have already been created to facilitate the SDC process involving digital sets of numerical or symbolic data. The monograph presents an overview of the most commonly used IT tools for SDC. The section starts with the presentation of two, probably most well-known, open source programmes: &#x1d70f;–Argus and &#x1d707;–Argus. Both were developed by Statistics Netherlands (Centraal Bureau voor de Statistiek) in the course of a few European projects. They are Java-based programmes, available with and without a bundled JRE7 distribution. Other SDC tools have been implemented in the R software and include a number of dedicated packages, such as sdcTable, sdcMicro, recordSwapping, cellKey, etc. The section ends with a brief description of possibilities offered by other SDC tools.

The protection of data confidentiality is also associated with organisational problems related to releasing official statistics and the need to check them to prevent a disclosure of sensitive information. The monograph presents arguments in favour of releasing statistical outputs, including appropriately prepared unit-level data for scientific research purposes (Scientific Use Files), different ways in which access to such files can be granted as well as specific requirements associated with such forms of release. These include formal requirements that persons or institutions applying for access to microdata must satisfy, as well as requirements regarding the level of protection that the data administrator (typically a NSI, but this can be any data holder) should guarantee in order to prevent unauthorised persons from gaining access to sensitive information.

The publication of the monograph was co-financed from the state budget under the “Excellent Science” programme of the Polish Minister of Education and Science and will be available both in electronic and printed form.

As already noted, statistical data are necessary to obtain reliable and valuable analytical outputs. It is worth noting that many students, when writing their theses (BSc., MSc., PhD) or other studies not only rely on available data from various sources but conduct their own polls, experiments or other surveys. Data collected in this polls may also contain sensitive information, which should be properly protected. Graduates can be also use data provided by official statistics or other custodians as part of their work. It is therefore important they should be aware of SDC rules, methods and tools and their impact on the final output.

It is with students’ needs in this respect that Poznań University of Economics and Business (PUEB) has included a special course about statistical disclosure control in the curriculum. Its aims are to familiarise students with key issues and concepts connected with the protection of data confidentiality, to present methods and techniques that help to maintain confidentiality of released data and to teach them how to use dedicated software when working with sensitive information.

After completing this course student should know the aims, principles and basic concepts of SDC, the main stages of the SDC process, specific SDC methods and techniques and should understand the trade-off between minimization of disclosure risk and information loss due to the application of SDC tools. They should also be able to identify potential threats relating to data confidentiality and choose optimal SDC methods and tools in specific situations.

The detailed syllabus of the course includes the following topics:

6

 protection of data confidentiality – aims and basic principles  definition of basic concepts  typology of output data  legal solutions and formal or ethical principles used in international practice regarding the protection

of data confidentiality  measurement and assessment of the risk of disclosure of confidential information depending on the

type of output data  non-perturbative and perturbative methods and techniques in SDC  information loss and its measurement  an overview of selected software used in the SDC process.

This lecture/course is conducted by experts in the field. The group of students expected to participate in the course and the number of lecturers may expand in the future.

It is also important that diploma thesis supervisors should pay special attention to the quality of data obtained by students and how it is affected by SDC and should instil in them the need to always protect sensitive information included in the collected data.

5. Concluding remarks

One of key tasks of modern statistics is to educate as many people as about the aims, rules and methods of SDC. A variety of measures are necessary to fulfil this task on the part of those who are involved in preparing and disseminating statistical data and those who use them. Initiatives and ideas presented in this article have already been implemented and can potentially reach many people, especially of these measures are further developed.

However, the proposed solutions do not exhaust the range of possibilities in this respect. One can think of other ways promoting SDC among various data holders, such as handbooks and guidelines disseminated among employees, organising internal trainings and lectures, providing assistance or consultations, if necessary. Currently, activities in this regard (apart from the SDC User Group within the Centre of Excellence) are still far from sufficient. It is therefore necessary to consider how to organise such cooperation and ensure it can be sustainable in terms of specialised staff involved and financing. Of course, particular organisations operate in different circumstances. While detailed regulations and mechanisms concerning data privacy depend on the sector and type of organisation, there are some common realities, principles and solutions which can be beneficial to all of them.

It is obvious that SDC topics in statistical education should be presented to university students, especially those whose fields of study depend on the collection and use of statistical data. However, pupils in secondary schools also learn fundamentals of statistics and conduct simple data analyses. It therefore reasonable that they, too, should become familiar with the most essential aspects of SDC. One way in which this can be achieved is by organising competitions, such as the European Statistics Competition or national competitions in statistics held in various countries (e.g. the annual Statistical Olympiad held in Poland since 2016). It would be good if such competitions included questions or subjects related to data protection and SDC.

6. References

Młodak, A., Pietrzak, M., Klimanek, T., Józefowski, T. & Lańduch, P. (2023), Confidentiality vs. utility of statistical information. Dilemmas of statistical disclosure control, Poznań University of Economics and Business Press, Poznań, Poland (in Polish).

Statistics Poland (2023), Guide for official statistics service units on the control of statistical data disclosure control, Collective study of the Team for Methods of Statistical Disclosure Control, Warsaw (under preparation).

SDC in statistical education – the Polish experience Tomasz Klimanek Tomasz Józefowski Andrzej Młodak

Statistical Office in Poznań, Poland

Poznań, September 2023

1

2

Introduction The growing demand for detailed statistical data requires the use and promotion of Statistical Disclosure Control (SDC). The staff of national statistical institutes (NSI) should have a broad knowledge of various SDC methods and tools, be able to apply them to obtain sufficiently safe and useful data to be released. Newly employed NSI staff could benefit from SDC training, which should include an overview of basis SDC problems. The knowledge of efficient ways of protecting the privacy of data stake- holders or customers is also important for many data custodians, who collect data to perform their activities and tasks. Users of statistical data (students preparing e.g. diploma theses, scien- tists, analysts, etc.) should also be aware of the benefits, drawbacks and expected effects of SDC methods. We will present various ways of disseminating knowledge about data privacy protection and SDC among these target groups that have alre- ady been applied or that will be introduced in the near future in Poland.

3

Outline of the presentation

1 Training of experts

2 SDC issues for new employees of official statistics

3 Education of potential data users and other stakeholders

4 Concluding remarks

5 References

4

Training of experts Basic assumptions

SDC education should start with NSI staff responsible for efficient data protection. The goal is to provide them with a broad knowledge of SDC and how these methods can be applied to survey data in order to maxi- mally protect the privacy of respondents while simultaneously ensuring the maximum utility of data for end users. Basic knowledge of SDC should also be disseminated among data users and taught to people who handle statistics, e.g. data managers in va- rious institutions and economic entities. In Poland, there is a dedicated team responsible for SDCmethods, which prepares the principles and guidelines on how to conduct the SDC pro- cess in various surveys. Moreover, units responsible for particular su- rveys should have their own methodologists responsible for SDC. To ensure that all NSI units in Poland have the required knowledge of SDC, a series of trainings have been conducted. They mainly took place in 2019 and were divided into basic and advanced courses.

5

Training of experts Basic training for NSI staff

General concepts and definitions Regulations and principles used in the international practice of SDC Basic types of disclosed information – tabular data and microdata Disclosure risk and information loss (disclosure scenarios; individual, global and hierarchical risk; measurement of risk; information loss due to the application of SDC; ways ofmeasuring information loss; the trade- off betweenminimizing disclosure risk andminimizing information loss) Mainmethods of protecting sensitive information in frequency andma- gnitude tables (non-perturbative and perturbative) SDC methods and techniques for microdata (non-perturbative and per- turbative) Software used for protecting statistical confidentiality (τ-Argus,µ-Argus, R tools, etc.) Problems connectedwith the preparation ofmicrodata for external users (public use files) – e.g. data from the LFS or EU-SILC.

6

Training of experts Advanced and other trainings within NSI

The advanced training included specific methodological and technical problems of SDC and practical exercises, i.e.:

application of dedicated R packages – e.g. sdcTable and sdcMicro, methods of generating synthetic microdata and new algorithms for protecting statistical confidentiality, e.g. the cell-key method, problems connected with the preparation of microdata for external users (PUFs), e.g. from the LFS or the EU-SILC, methods of protecting statistical confidentiality of census data.

The third training featuring similar issues but in amore condensed form was addressed to members of the expert team, who were planning to use the presented methods and tools in their daily work but faced pro- blems such as insufficient experience in the use of R, the lack of com- petent and experienced staff in the use of SDC. Recently SDC training has been offered in the form of direct individual consultations or on-line sessions. The Team for Methods of Statistical Disclosure Control is preparing special guidelines on SDC for NSI em- ployees (Statistics Poland (2023)).

7

SDC issues for new employees of official statistics SDC in preparatory service

Preparatory service training for people employed in public administra- tion includes learning about basic mechanisms and rules that govern specific public agencies and consists of two parts: obligatory (divided into basic and extended) and optional Since 2021, the methodological aspects of SDC have been covered as part of topic 3 (principles of organising statistical surveys). The section Confidentiality and utility of output data in official statistics includes:

the legal basis of statistical confidentiality anonymisation and pseudonymisation aims and basic concepts of SDC disclosure, disclosure risk and information loss user as an analyst and user as an intruder the SDC process, its components and stages SDC for microdata, tabular data and statistical outputs.

The effectiveness of the training is evaluated by means of a quiz to check if the knowledge has been properly acquired.

8

Education of potential data users and other stakeholders A monograph aimed at popularizing SDC

The knowledge of SDC needs to be broadly disseminated not only among NSI staff but also among various data holders responsible for the pro- tection of data confidentiality, users conducting analyses based on sta- tistical data as well as people who collect any kind of sensitive data as part of their own research. In an effort to make such knowledge easily accessible, a group of NSI employees from the Statistical Office in Poznań have prepared a hand- book (Młodak et al., 2023). The monograph is an attempt to provide a comprehensive description of all aspects of SDC, including the goal and definition of statistical disclosure control, its formal and legal principles and types of released data (including metadata, paradata and other data). The disclosure process, types of data users, typologies of statistical outputs with respect to the protection of sensitive information and the trade-off between disclosure risk and data utility are explained.

9

Education of potential data users and other stakeholders A monograph aimed at popularizing SDC

Themonograph describes SDCmethods formicrodata, tabular data and output checking, presenting characteristics of each type of data, their sources and fundamental principles of their protection, mathematical methods and IT tools used for this purpose as well as organisational and technological aspects of releasing statistical data, which are asso- ciated with the risk of unit re-identification or disclosure of attribute. The main purpose of SDC is to achieve an optimal trade-off between minimisation of disclosure risk and information loss. Therefore, the au- thors of the monograph pay a lot of attention to the problems of me- asuring these quantities (in the case of disclosure risk – in the context of internal and external risk) Themonograph provides numerous examples showing how to construct such measures and how to interpret them; the examples demonstrate the usefulness of various measures for different use cases of released data, including measures of estimation precision.

10

Education of potential data users and other stakeholders A monograph aimed at popularizing SDC

Fig. 1: The monograph’s cover page (in Polish) & the first page of the summary (in English)

11

Education of potential data users and other stakeholders Support for students

It is worth noting that many students, when writing their theses (BSc., MSc., PhD) or other studies not only rely on available data from various sources but conduct their own polls, experiments or other surveys. Data collected in these polls may also contain sensitive information, which should be properly protected. Graduates can also use data provided by official statistics or other cu- stodians as part of their work. Poznań University of Economics and Business (PUEB) has included a special course about statistical disclosure control in the curriculum (to start in 2024/2025). Its aims are:

to familiarise students with key issues and concepts connected with the protection of data confidentiality to present methods and techniques that help to maintain confidentiality of released data to teach them how to use dedicated software when working with sensitive information.

12

Education of potential data users and other stakeholders Support for students

The detailed syllabus of the course includes the following topics: protection of data confidentiality – aims and basic principles definition of basic concepts typology of output data legal solutions and formal or ethical principles used in international practice regarding the protection of data confidentiality measurement and assessment of the risk of disclosure of confidential information depending on the type of output data non-perturbative and perturbative methods and techniques in SDC information loss and its measurement an overview of selected software used in the SDC process.

This lecture/course is conducted by experts in the field. The training coverage may expand in the future. Diploma thesis supervisors should pay special attention to the quality of data obtained by students and how it is affected by SDC and instil in them the need to protect sensitive information included in the data.

13

Concluding remarks Concluding remarks

One of key tasks of modern statistics is to educate as many people as possible about the aims, rules and variety of SDC methods. Initiatives and ideas presented so far have already been implemented and can potentially reach many people. One can also think of other ways of promoting SDC among various data holders, such as handbooks and guidelines disseminated among em- ployees, organising internal trainings and lectures, providing assistance or consultations, if necessary. Currently, activities in this regard (apart from the SDC User Group within the Centre of Excellence) are still far from sufficient. Students’ education in SDC is very important. However, pupils in secon- dary schools also learn fundamentals of statistics and conduct simple data analyses and hence they should also become basically familiar with it. One way in which this can be achieved is by organising compe- titions, such as the European Statistics Competition or national compe- titions in statistics held in various countries (e.g. the annual Statistical Olympiad held in Poland since 2016). It would be good if such compe- titions included problems related to data protection and SDC.

14

References References

Młodak, A., Pietrzak, M., Klimanek, T., Józefowski, T. & Lańduch, P. (2023), Confi- dentiality vs. utility of statistical information. Dilemmas of statistical disclosure control, Poznań University of Economics and Business Press, Poznań, Poland (in Polish). Statistics Poland (2023), Guide for official statistics service units on the control of statistical data disclosure control, Collective study of the Team for Methods of Statistical Disclosure Control, Warsaw (under preparation).

Thank you very much for your attention!

15