Issue, No.22 (June 2022)

France is back: a new series to help understand recent and old French dynamics of inequalities (1970-2018)

by Louis Chauvel (University of Luxembourg)

With the new LIS data release, France is part once more of comparative inequality studies: for many years, the available French datasets only ranged from 1978 to 2010 and there was no clear indication as to when this out of date series might be refreshed. This wait is now over and a new series covers the period 1970-2018, with the promise of yearly updates from here on.

The tale of two surveys

In brief, France only held two main series of periodic surveys with potentially available microdata: the “old” Enquête Budget des ménages (previously “des familles”), aka Household Budget Surveys (HBS), and the “new” ERF or ERFS series of Enquête sur les revenus fiscaux (previously “et sociaux”), aka Tax Income Survey (TIS). In a nutshell:

  • The old HBS is a pervasive consumption survey of circa 10.000 households per release, taken every 5 years approximately: 1978, 1984, 1989, 1994, 2000, 2005, 2010. On top of systematic information regarding expenditure over 14 days, it also collects complete income components compatible with the LIS template for the previous 12 months. This means an additional two-hundred thousand individuals in the LIS collection.
  • The new TIS is based on larger samples (up to 50.000 households) and a smaller set of variables. TIS has been run yearly from 1996 on and though it is primarily a “tax” survey with administrative data, it is backed by the Enquête Emploi, a hybrid between a Labor Force Survey and a Current Population Survey, including a complete set of demographic and economic indicators, meaning that a large amount of information, including education and migration details, amongst others, are available. TIS means an overall sample of three million individuals aggregated from 1970.

From the old HBS to the new TIS, one might expect more precise, reliable sampling, faster availability for the recent years, and officially checked (taxable income and validated social rights) variables. An important aspect of yearly surveys is not only the rapid availability of fresh information, but also the capacity to detect discontinuities and outliers in the series. In this instance the main gap is that, before 1996, basic socioeconomic variables such as education, property ownership or migration, etc., were not collected in the TIS but were present in the HBS.

Comparing the two series

The best way to evaluate the added value of the new series is to compare it with the old one, in relation to other available references on French inequalities, including the most recent 2021 picture provided by the French official institute of statistics Insee (Insee, 2021, Les revenus et le patrimoine des ménages, Edition 2021, Paris, Insee).

Detection of outliers: the FR1984 HBS Survey exception

Even if the Gini index is not a panacea in income inequality analysis, it deserves some attention. It is relevant to compare the two series in these terms. Before the availability of the TIS series, there were concerns about the reliability of FR1984 HBS since it was three percentage points of the Gini index above the linear trend for 1978-1989; though some arguments (capital gains at the top, new poverty at the bottom) were put forward, not one was convincing. With the TIS series as a benchmark, it is clear that FR1984 HBS was an outlier overshooting the true Gini by perhaps as much as four points.

After a rectification (see fig 2) of 1984, based on a lower weighting at the bottom and top tails of the distribution, the long-term shape of the French Gini 1970-2018 showed as a U-curve, with a minimum in the 1990s and a local maximum in 2010. The (modest) increase of income inequality between the end of the 1990s and 2018 is confirmed in the literature (Insee, 2021, p.18).


Fig 1: Comparison between the old HBS (dots) and new TIS (lines) series for gini_dhi

Source: Luxembourg Income Study (LIS) Database.

Comparison of macro indicators of inequality: advantage of TIS over HBS

Beyond the Gini index, we compare a set of indicators to understand the proximities and gaps between the HBS and TIS series. First, the averaged logged disposable income in euros 2020 (ley) shows the great slowdown of economic growth after the 1970s. The HBS trend is globally similar, but fails to detect its details, like the economic acceleration of the 2000s and the great recession of the post 2008 era, details that the TIS surveys clearly bring out. The two main added values of the TIS compared to the HBS are the possibility to go back to 1970 and forward until 2018.


Fig 2: Comparison between the old HBS (dots) and new TIS (lines) series for six indicators (FR1984_HBS rectified):
ley: Logged equivalized disposable income in real terms (eur2020)
gini_dhi: Gini index of the equivalized disposable income
p50: relative poverty rate (50%), proportion of the population below half (50%) the median equivalized disposable income
r200: relative rich rate (200%), proportion of the population above twice (200%) the median equivalized disposable income
sh_bottom50: share of total equivalized disposable income of the bottom 50% of the population
sh_top10: share of total equivalized disposable income of the top 10% of the population

Source: Luxembourg Income Study (LIS) Database.

We also compare four additional indicators, two regarding the bottom of the distribution and two for the top. For the lower part, we have the traditional poverty rate at 50% of the median, and the share of the total income share of the lower 50% of the population. The poverty rate in the TIS clearly shows a floor in the 1990s, when minimum universal income was implemented for the first time. The HBS tends to confirm a similar V curve, but with a somewhat less credible floor in 2000. The aggregated share of the proportion below the median population is less informative, with a value close to 30.5% +/- 1% in the two sources.

Symmetrically, at the top, we follow and compare the proportion of “rich” people, and the share of the top 10%. The “relative richness rate” at 200% of the median, symmetrical to the relative poverty rate, is an indicator that French inequality specialists particularly appreciate (Chauvel L., 1995, « Inégalités singulières et plurielles : l’évolution de la courbe de répartition des revenus », Revue de l’OFCE, 55, pp. 211-240; A.Brunner et L.Maurin (eds.), 2022, Rapport sur les riches en France, édition 2022, Observatoire des inégalités, juin 2022, see p.14). The TIS series show the strong decline in richness rate in the 1970s from almost 12% to around 8% in 1978, where modest fluctuations might be observed near to a plateau at 8%. More interesting is the share of the top 10%, with a shaky V curve with a floor in the late 1990s. Once again, the TIS source provides details that HBS cannot deliver.

The overall diagnosis is that for macro indicators of inequality, even after a rectification of the 1984-HBS that caused a major problem, the TIS is better than HBS, as regards the size of its samples, the frequency of information, the continuity of results from year to year, and therefore general reliability. For the computation of key figures in LIS-, TIS is preferable.


Structural changes in the long range: equality HBS-TIS

With respect to fine-grained LIS analysis, including controls for important covariates (education, region, migration, etc.), the HBS-TIS match becomes complicated: the TIS template and codebook is relatively complete from 1996 onwards, where TIS is preferable to the end of the HBS series (the four HBS surveys from 1995 to 2010). However, before 1996, the TIS surveys lack basic information, vital when one wishes to have accurate socioeconomic controls. In this respect, HBS 1978-1995 may have a major role to play in recording two additional decades of historical changes.

The solution to the TIS issue (education is absent from TIS before 1996) is backed by empirical facts: in terms of education, TIS after 1996 and HBS provide parallel observations. The HBS series for 1978-1994 provide useful information: the U-curve of risk of poverty for the least educated population across the period 1978-2018, and the declining opportunities of richness for the most educated ones, due to overcrowding effects (or “diploma inflation”: see Collins R. (2019), The Credential Society : An Historical Sociology of Education and Stratification (New Preface), Legacy Editions.. The trend of a decline in the return to higher education in terms of a proportion more than twice the median income begins in the 1990s: prior to this date, the HBS series show a plateau.


Fig 3: Comparison between the old HBS (dots) and new TIS (lines) series for poverty rate (left) and rich rate (right) by level of education:
highly educated (15 years of education or more: triangles and grey lines)
averagely educated (11 to 14 years of education, circles and blue lines)
low-educated (10 years of education or less, crosses and green lines)
(FR1984_HBS rectified)

Source: Luxembourg Income Study (LIS) Database.

Conclusion: TIS is better for the present and HBS remains vital for the past

All in all, this HBS/TIS comparison is an opportunity to evaluate the relative merits of the different series. For macro analyses up to the present day, TIS outperforms HBS. For controls, TIS cannot replace HBS before 1996, due to a lack of strategic covariates such as education and other socioeconomic variables in the TIS: HBS microdata has to remain in the LIS archives.

In this attempt to evaluate the many “pros” and few “cons” of TIS, my personal conclusion is inspired by Paul Feyerabend’s epistemology (P. Feyerabend, Against Method: Outline of an Anarchistic Theory of Knowledge, London: New Left Books, 197): “anything goes”! Imperfectly translated in French from “tout est bon”. More accurately, “fromage ET dessert” is what we deserve: HBS AND TIS.