Measuring the Size of the “Third Pillar”: A Global Dataset

Abstract

Balancing the state and the market, the “third pillar” (i.e., any association of people that is neither public nor private) is a key sector in society. In contrast to the first two pillars (the public sector and the private sector), the third pillar has received very little attention in academic and policy debates. This paper aims to facilitate research on the third pillar's relevance by constructing the first global dataset on the size of the third pillar, including estimates for 120 countries. Results show that the size of the third pillar is substantial, both when measured as third pillar time (TPT) or third pillar participation (TPP). Global TPT is equivalent to 5.1 percent of total employment, more than half of the global agricultural sector. Moreover, findings indicate that TPP equals 13.4 percent, suggesting that more than one person in eight of the world's adult population is active as a volunteer in the third pillar.

1. Introduction

The size and relevance of the state and the market is widely documented, with frequently cited reports such as the OECD's “Government at a Glance,” or the World Bank's “Ease of doing business” ranking. However, the “third pillar” of society has been overshadowed by the ideological and political polarization between giving more importance to markets and to the state, which has increased over the past decades. Mintzberg (2015) defined the third pillar as “any association of people that is neither public nor private—owned neither by the state nor by private investors. Some are owned by their members; others are owned by no one.”¹Mintzberg's (2015) definition captures the variety of associations and ownerships included within the third pillar, ranging from cooperatives and Red Cross national societies to activist associations. Throughout this paper, the term “third pillar” is used as an umbrella term for other labels such as “social economy”, “nonprofits”, “civil society”, or “third sector.” Rajan (2019) argued in favor of strengthening local communities, the third pillar of society, as an antidote to growing despair and unrest around the globe. He further argued that in balancing the state (providing protection) and the market (providing consumption), the third pillar shields society both from the dangers of a collectivist state and from individualism.

The third pillar is underestimated in both size and relevance for two reasons. First, the absence of standardized data currently does not allow for benchmarking across countries or over time. Although frameworks to measure (segments of) the third pillar are available, they are not widely implemented by national statistical authorities. Second, methodological challenges inhibit reliable estimates of the “added value” of the third pillar. While the issues to determine the value of government services (Kemp and Burt 2001) can be—to some extent—overcome by valuation at cost, this is not a possibility for the third pillar. Voluntary contributions and nonprofit objectives are core elements of the third pillar, which are arguably valuable and yet cannot be valuated at cost.

In this study, the absence of standardized data is addressed by constructing a global dataset that includes 120 countries, more than doubling the current number of countries with available data. This methodology yields highly accurate predictions of third pillar size: the model is able to explain 78 percent of the variation in actual size of the third pillar in the 51 countries with complete data. The Random Forest model leverages available data on the structure of the labor market, the level of economic development, and the political regime, without imposing a functional form. However, the range between 5^th and 95^th percentiles (obtained via Jackknife procedure) indicates noisy estimates. Essentially, the quality of the estimates depends on the quality of existing third pillar size data. Therefore, standardized data collection—using existing frameworks—leading to reliable and more broadly available data will be needed to further improve the dataset presented here. Furthermore, it should be acknowledged that there is a need for additional research about the added value of the third pillar. As both are interrelated, available standardized data on the size of the third pillar could also facilitate empirical research on its added value. By constructing a global dataset, this will facilitate the integration of the third pillar into academic discussions, offering the potential for valuable insights to researchers and supporting the re-evaluation of the prevailing perspectives on the three pillars within contemporary society. Additionally, understanding the size of the third pillar in each country carries significant policy implications, particularly considering the tendency of non-governmental organizations (NGOs) to focus only on local and closely related entities in their campaigns (Hatte and Koenig 2020). A global dataset offers policy makers a “macro” perspective on the third pillar and helps them target policy initiatives. While interest in the third pillar—proxied by Google Trends data—appears to be in decline since 2009, the activity related to the third pillar—proxied by the share of the population freeing up time (volunteering) or savings (donating money)—appears to have increased (fig. 1). This discrepancy might be explained by the absence of academic research on the third pillar to inform the public debate, which has become increasingly polarized between giving more emphasis either to markets or to the state.

The remainder of the paper is structured as follows. Section 2 defines the third pillar and discusses existing measurement frameworks and available datasets. Section 3 summarizes the data used to construct the global dataset and further explains the methodology to predict the size of the third pillar. Moreover, the predicted values are presented and stylized facts about the size of the third pillar are discussed. Section 4 formulates conclusions and discusses potential avenues for future research.

2. Measuring the Third Pillar

Definition and Measurement

The most common approach to measure the size of the third pillar is to use a “time-based” proxy (e.g., Salamon and Sokolowski 2018). Two time-based proxies for third pillar size can be distinguished: third pillar time (TPT) and third pillar participation (TPP). TPT indicates the time spent in the third pillar as a percentage of total employment, including both paid staff (full-time equivalent; FTE) and volunteers in organizations (not including “direct” volunteering, see below), converted to (a portion of) FTE based on their time spent volunteering. TPP indicates the share of the population active as a volunteer (e.g., fig. 1). The advantage of TPT over TPP is that it enables a comparison of third pillar size not only across countries, but also relative to other economic sectors (public or private). As will be described in more detail below, TPT was used as the main proxy for third pillar size, but an estimate for TPP was calculated as well.

An alternative approach to measure the size of the third pillar is a “value-based” proxy, which estimates the (gross) “value-added” of the nonprofit sector, with the advantage of having the ability to integrate these monetary estimates into the national accounts. Examples of countries that have adopted this approach include Mexico, New Zealand, Mozambique, Norway, and Thailand. However, the value-based approach follows a market-based logic detailing revenues and costs. Following this logic, the value-based approach fails to capture the value of unpaid volunteering, unless a proxy for the value of time or the added value of the end product (rather than the unpaid volunteering itself) is used.² One example of a method used to estimate this proxy is calculating the social return on investment (SROI). However, causal inference is needed to reliably estimate the SROI. For many third pillar activities where services are provided (e.g., free first aid where courses are taught by volunteers), there are no data available to obtain these causal estimates. Even when the valuation of end products is possible, reducing the volunteering component in third pillar activities can counterintuitively increase the value-added of this sector. For example, following the market logic, voluntary non-remunerated blood donations are valued at their unit price when blood products are sold to hospitals or pharmaceutical companies at cost. However, the time spent by voluntary blood donors—and even their blood—is not valued. If one would shift the system towards paid donations, the value-added calculation of the blood banks would increase, as prices would have to incorporate the required fees paid to donors. However, if this shift from voluntary to paid donations occurred, it is rather unlikely that the actual added value of this activity would increase, considering the risks of paid donations for donors (e.g., Zhao et al. 2021; MacKay, Schiltz, and Vandekerckhove 2024) and the risks for patients receiving the blood products (e.g., World Health Organization 2010). Paradoxically, and further exemplifying the issue with value-based proxies for third pillar size, these risks would in turn necessitate additional lab testing, resulting in an even more “increased” value-added calculation for paid donations.

In this paper, the time-based approach of measuring the size of the third pillar is used, not only to compare estimates with the literature as it is the most common approach (e.g., Salamon and Sokolowski 2018) but also given the issues with accurately measuring the added value of activities in the third pillar using a value-based approach. In section 3, TPT is calculated using a modeling approach and a back-of-the-envelope estimate is calculated for TPP. The exact definitions of the third pillar used below are based on existing datasets from international surveys and frameworks (Salamon, Sokolowski, and Haddock 2017; Salamon and Sokolowski 2018), rather than creating a renewed definition of what constitutes volunteering or third pillar organizations.

Available Frameworks and Datasets

Different frameworks are available to measure the size of the third pillar, yet adoption of these frameworks remains low. Since its release in 2003, only a handful of countries have implemented suggestions from the Handbook on Nonprofit Institutions in the System of National Accounts (United Nations Statistics Division 2003). Data on volunteering are more widely available with the International Labour Organization's (ILO) Manual on the Measurement of Volunteer Work (International Labour Organization 2011) as the reference approach. A more recent edition provides an add-on module on volunteer work in national labor force surveys, enabling the potential for standardized data collection across countries (International Labour Organization 2021). The ILO distinguishes between “direct” and “organization-based volunteering.”³ Between 2010 and 2021, ILO collected data to construct a volunteer rate (percent of working-age population) for 52 countries.

In their 2018 handbook, the UN incorporated the ILO standards in its Satellite Account on Nonprofit and Related Institutions and Volunteer Work (United Nations 2018). This framework can be considered the "gold standard" for the measurement of the third pillar, and includes third pillar estimates from different angles as described above (i.e., time-based and value-based).⁴ However, since its publication in 2018, the handbook has reached less than 70 downloads.⁵

The Johns Hopkins Comparative Nonprofit Sector Project (Johns Hopkins University [JHU]) generated international data for 44 countries (Salamon, Sokolowski, and Haddock 2017) and further conducted a specific study on the third pillar for the 28 countries in the European Union (at the time of publication still including the United Kingdom) and Norway (Salamon and Sokolowski 2018). The international JHU dataset includes both paid staff and volunteers (converted to FTE) to measure TPT. In addition, the EU JHU dataset includes “direct” volunteering as defined above.

The variety of frameworks, the low adoption rates of existing frameworks by countries, and the scattered data complicate empirical academic research on the size and relevance of the third pillar. As a result, the third pillar receives little attention, thereby making the adoption of existing frameworks a low priority for policy makers. In order to break this vicious cycle, this study proposes a novel methodology to construct a global dataset thereby facilitating further empirical research on the relevance of the third pillar.

3. Constructing a Global Dataset

Data

Five sources were combined into one dataset as a basis for the predictive model (see “Methodology”): the international JHU dataset, the EU JHU dataset, the ILO labour market dataset, the World Development Indicators dataset, and the Polity5 Project dataset. After combining all five sources, 51 countries with complete data remained (i.e., data on working-age population, participation in workforce, labor-market structure, level of economic development, political regime, and third pillar size).⁶

The size of the third pillar (table 1, panel 5) was used as the predicted variable and was measured as TPT.⁷ When TPT was available for a country in both the EU and the international JHU dataset, the estimates of the EU JHU dataset were retained, as these estimates were more recent compared to the international ones.⁸

Three main sets of predictor variables were selected: the labor-market structure, the level of economic development, and the political regime. The labor-market structure was included (see table 1, panels 1 and 2) because it could be a predictor of TPT. Specifically, both the activity within the labor market (i.e., participation, employment) and the sector of employment (e.g., agriculture, trade) could correlate with TPT (either as volunteer or paid employment). For example, countries with a lower participation rate could be characterized by more volunteering. The ILO labour market dataset was used to obtain country-level information on the structure of the labor market, and eight predictors were selected: the size of the population (aged 15 to 64), the participation rate, and the relative employment for six sectors in the economy: Agriculture; Manufacturing; Construction; Trade (trade, transportation, accommodation and food, business, and administrative services); Public (public administration, community, social and other services and activities); and Other⁹ (mining and quarrying, electricity, gas, water supply, and not classified).

The economic development of a country was included as a predictor (GDP per capita; table 1, panel 3) because it could correlate with TPT in that country. For example, individuals with a higher standard of living would potentially have more flexibility to free up time for volunteering. Furthermore, Dupuy, Ron, and Prakash (2016) showed that low- and middle-income countries implement restrictive laws to receive less foreign aid for domestic NGOs in order to limit potential political competition. This suggests an alternative mechanism that could explain a positive correlation between GDP per capita and TPT. World Development Indicators from the World Bank were used to obtain country-level information about the level of economic development for the global dataset.¹⁰

Indicators related to the political regime (table 1, panel 4) were included in the global dataset as a third set of predictor variables. Specifically, three predictors were chosen: Polity index, democracy, and durability. The political regime could predict TPT if, for example, the third pillar develops faster and more easily in more stable and democratic societies than in autocratic regimes, in which the third pillar might be seen as political competition. Information about the political regime in countries was collected via the Polity5 dataset.

Summary statistics are presented in table 1. On average, included countries have a working-age population of just under 50 million people, and over 70 percent workforce participation. The trade sector is the largest employer on average, followed by the public sector and manufacturing. Country-level characteristics (economy and democracy) vary widely, indicating that the dataset represents countries in different income groups and democratic institutions. Time spent in the third pillar (TPT) represents on average 7 percent of total employment, for countries where data are available. This is mostly due to paid employment (4.5 percent), which is intuitive: assuming a 40-hour work week, 20 volunteers with an average activity of 2 hours per week are needed to account for 1 FTE.

Methodology

To estimate TPT, a predictive model was trained on the dataset composed of the 51 countries with complete data. Subsequently, this model was applied to predict TPT for the remaining 69 countries, for which data of all predictors were complete (i.e., labor-market structure, level of economic development, and political regime), but lacked information on TPT.

When predicting TPT, a Random Forest model was employed in order to allow for maximal flexibility in linking the predictors (i.e., the structure of the labor market, the level of economic development, and the characteristics of the political regime) to the predicted variable (i.e., the size of the third pillar; TPT). This machine-learning approach ensured that any potential interactions were captured between sectors to predict TPT, without imposing a (linear) functional form. Machine-learning models were selected, as they tend to have superior performance when the goal is predictive power as opposed to causal inference (James et al. 2013; Varian 2014; Ludwig and Mullainathan 2021).¹¹

To estimate the accuracy of the TPT prediction, the sample was divided into a training set and a test set. The training set was then used to estimate the parameters of the model, and the test set was used to determine the accuracy of the model based on the parameters estimated on the training set. If the model was instead trained and tested on the same dataset, the precision of the model would likely have been overestimated (James et al. 2013). Instead of using a single split of the data into a training and test set, the k-Fold Cross-Validation (CV) method was used to limit the influence of one particular sample split. This method randomly divided the observations into k subsets (folds). Therefore, it fit k models, where the k^th model was estimated based on all data except the k^th fold. The k^th model was then used to make predictions for all data in the k^th fold. In this way, an “honest” prediction was obtained for each observation. As such, a Random Forest model was used along with the CV method (using k = 10) of the SuperLearner package in R. Both the mean squared error and R-squared are reported. Subsequently, the Jackknife method developed by Wager, Hastie, and Efron (2014) was used to obtain the 5^th and 95^th percentiles of the TPT estimate.

Separate models were estimated for TPT including both paid staff and volunteers (“total”), and including only paid staff (“paid”).¹² The predictive performance of both models is displayed in table 2. The average of the MSE for the “total” TPT model across the 10 models was equal to 6.18 (4.20 for the “paid” TPT model). On average, across the 10 models, the model explained 78.29 percent of the observed variation in “total” TPT (78.31 percent for the “paid” TPT model). Hence, using information on the structure of the labor market, the level of economic development, and the political regime, more than three-fourths of the variation in TPT can be explained. In fig. 2, a scatterplot of the predicted and the actual values for both TPT measurements (i.e., “total” and “paid”) can be found. The figure shows that for both models, a large portion of the variation in TPT is explained by the predictions of the model, as indicated by the limited deviations from the 45° line (i.e., representing a perfect correlation).

Results

Table 3 and fig. 3 summarize the results of the 120 country-level estimates of TPT. The available TPT estimates for 51 countries are supplemented by predictions for an additional 69 countries for which there was only information on the structure of the labor market, the level of economic development, and the political regime, with no information on TPT. Table S1.1 in the supplementary online appendix lists the actual TPT (if available), the average predicted TPT, the 5^th and 95^th percentiles of the predicted TPT, and the estimated split of TPT into paid staff and volunteers. As can be seen from this table, actual TPT for the 51 countries with available data consistently lies within the 5^th–95^th percentile range of the predictions, confirming the predictive power of the model.

On average, based on the model, TPT accounts for 5.1 percent of total employment, equivalent to more than half of employment in the global agricultural sector (9.7 percent). The Netherlands (17.4 percent), Luxembourg (17.0 percent), and Belgium (14.7 percent) are the countries with the largest TPT, whereas Romania (1.0 percent), Pakistan (1.0 percent), and Lithuania (1.1 percent) have the smallest TPT. A breakdown by regions and subregions (ILO definitions) is shown in table 3, and global variation is visualized in fig. 3. While Eastern Europe (2.5 percent) and Southern Asia (2.0 percent) have a relatively small TPT, Northern America (which includes Bermuda, Canada, Greenland, Saint Pierre and Miquelon, and United States if data were available) has a relatively large TPT (11.5 percent).

When grouping countries according to the World Bank's income groups, it is clear that high-income countries have the largest third pillar (8.7 percent), while variation between low-income (3.2 percent), lower-middle-income (3.0 percent) and upper-middle-income (3.7 percent) countries is rather limited. At a country level, there is no one-to-one relationship between the size of the economy and the size of the third pillar (TPT).¹³ In fig. 4, this can be seen from the variation in TPT across countries with similar GDP per capita. For example, Greece has a third pillar that is almost four times the size than that of Czech Republic, while their GDP per capita levels are comparable.

When decomposing Global TPT from table 3, the 5.1 percent consists mainly of paid staff hours (∼60 percent of total TPT), with volunteers contributing around 2 percent (∼40 percent of total TPT). However, these values represent the percentage of total employment, and therefore do not imply that only 1 in 50 persons are active as a volunteer in the third pillar. A better proxy of the number of volunteers is the participation of volunteers in the third pillar (TPP). The methodology described in Appendix B of the State of the World's Volunteerism Report (United Nations Volunteers Programme 2021) was employed, which involved computing a ratio between TPT and participation of volunteers, for countries that had data on both. For the latter, the ILO's indicator on organization-based volunteering was used, defined as the share of adults that voluntarily contributed in the past 12 months to an organization, community, or group (i.e., as part of the third pillar). Therefore, to calculate this ratio for the 36 countries with complete data, the ratio between the ILO estimate of organization-based volunteering and the estimate of time spent by volunteers in the third pillar (“volunteer TPT”) was computed.

On average, 1 percent of volunteer TPT was found to be equivalent to 7.1 percent of adults being active as a volunteer in third pillar organizations in the past 12 months. Applying this ratio to the country-level estimates of volunteer TPT, global TPP was found to be equal to 13.4 percent. Hence, this back-of-the-envelope computation suggests that more than one in eight of the world's adult population has been active as a volunteer¹⁴ in the third pillar in the past 12 months. Similar to TPT, there is broad variation in TPP across countries, with the lowest number of volunteers active in the third pillar in Egypt (0.75 percent), Turkey (1.6 percent), and South Africa (1.7 percent), and the highest number of volunteers in The Netherlands (40.3 percent), Canada (44.0 percent), and Norway (48.0 percent).

4. Discussion and Conclusion

The aim of this paper was to create a global dataset to facilitate further empirical research on the third pillar. Using available country-level data on the labor-market structure, economic development, and political regime, this study constructed a dataset on the size of the third pillar for 120 countries. Using this dataset, future research should aim to refine and complement the third pillar size estimates and to investigate the relationship between the size of the third pillar and different societal outcomes on a global level. One example of this relationship is the observation that countries with a larger third pillar—proxied by the size of societal volunteering—have been less impacted by COVID-19 (Schiltz et al. 2023). Another example is the study of Enjolras and Sivesind (2018) that describes positive associations between the size of the third pillar—measured as share of the third pillar's workforce in the total workforce—and life satisfaction, health, political engagement, and social trust. However, the statistical power in both of these studies is limited by the number of observations as well as the proxies employed for the size of the third pillar. The global dataset on the size of the third pillar introduced in this paper establishes a unified proxy for a broad set of countries, enabling further empirical research.

The current dataset is still imperfect, and a few limitations remain. First, the set of countries with reliable and comparable estimates for the size of the third pillar is too small at present. A broader adoption of the UN's Satellite Account on Nonprofit and Related Institutions and Volunteer Work is needed to remedy this issue. National statistical authorities should adopt the systematic inclusion of data on the third pillar in their national accounts, following the lead of the 18 countries who have done this already. A first step could be for countries to include the add-on module of the ILO, which has surveys on volunteer work (International Labour Organization 2021), thereby enabling them to gather standardized data on the size of societal volunteering, a key element of the third pillar.

Second, methodological issues limit the accuracy of TPT predictions. In the EU JHU dataset, the size of the third pillar for 10 countries¹⁵ was estimated using imputations due to a lack of data. Inserting these imputed values as a basis for the imputations in the global dataset might have inflated the bias in the estimates on the size of the third pillar.¹⁶ Comparing the size of the third pillar (5.1 percent) and the mean absolute error (i.e., square root of MSE; 2.5 percent) for the countries used to build the predictive model suggests that the estimates are noisy but still provide valuable information, especially when comparing the size of the third pillar across countries. To improve the global dataset, this noise could be further reduced by building a more comprehensive model with more predictors (at the cost of including fewer countries with available data) and by collecting more data on the third pillar. As more countries collect (standardized) information on the size of their third pillar, the predictions for the remaining countries will improve over time, which emphasizes the need for more systematic data collection.

Third, the definition used to measure the size of the third pillar should be refined further. Given the data constraints, a time-based - as opposed to a value-based - proxy (i.e., TPT) was argued to be a better choice for measuring the size of the third pillar. However, the back-of-the-envelope calculation for TPP highlighted that more than one in eight of the world's adult population has been active as a volunteer in the past 12 months in third pillar organizations, which seriously underestimates their activity when third pillar size is measured as TPT.

Contribution Statement

Philippe Vandekerckhove conceived of the presented idea and supervised the work (Conceptualization, Supervision). Fritz Schiltz collected the data and performed the computations (Methodology, Formal Analysis, Writing—Original Draft). Kelsey J. MacKay contributed to the analysis and writing (Formal Analysis, Figures and Tables, Writing—Revising). All authors discussed the results and contributed to the final manuscript (Writing—Review and Editing).

Conflict of Interest

All authors are employed by Belgian Red Cross–Flanders, responsible and reimbursed for supplying adequate quantities of safe blood products to hospitals in Flanders and Brussels.

Funding

This work was supported by the Belgian Red Cross Research Foundation.

Data Availability Statement

The data that support the findings of this study are publicly available at the following Github repository: Belgian Red Cross–Flanders, “Third Pillar Global Dataset,” https://github.com/Belgian-Red-Cross-Flanders/Third-Pillar-Global-Dataset.

Ethics Approval Statement

This study does not involve human participants or animal subjects.

Footnotes

Author Biography

Fritz Schiltz (corresponding author) is the director of policy and projects at Belgian Red Cross-Flanders, Mechelen, Belgium Motstraat 42, 2800 Mechelen, Belgium; his email address is [email protected]. Kelsey J. MacKay is a researcher at the Third Pillar Research Centre, Belgian Red Cross-Flanders, Mechelen, Belgium, Motstraat 42, 2800 Mechelen, Belgium; her email address is [email protected]. Philippe Vandekerckhove is the CEO of Belgian Red Cross–Flanders, Mechelen, Belgium; a professor at the Department of Public Health and Primary Care, Leuven Institute for Healthcare Policy, KU Leuven, Leuven, Belgium; a professor at the Centre for Evidence-Based Health Care, Stellenbosch University, Cape Town, South Africa; his email address is [email protected].

References