Do long constitutions really hamper economic performance? A comment on Tsebelis and Nardi (2016a)

Abstract

In their contribution, Tsebelis and Nardi (2016a) examine the impact of constitutional design on economic performance. The authors find evidence of a significant negative relationship between constitutional length and economic performance. As we will show here, the results of Tsebelis and Nardi (2016a) are not robust. Several minor adjustments to the models, each on its own, reduce or eliminate the presumed relevance of constitutional length. Moreover, conceptual considerations raise doubts whether the focus on constitutional length is a promising path to follow.

Introduction

One of the most important questions in politics and society is what determines economic success in a country. Accordingly, there is extensive research on this topic, particularly concerning the influence of political institutions (Knack and Keefer, 1995; North, 1990; Rodrik, 2004). In their significant and innovative contribution, Tsebelis and Nardi (2016a, hereafter Tsebelis and Nardi, 2016b) examine the impact of constitutional design on economic performance, thus contributing both to the research on institutions and economic success and on the field analyzing whether de jure (i.e., written) constitutions matter (e.g., Bologna Pavlik et al., 2023; Callais and Young, 2021; Chilton and Versteeg, 2016, 2017, Voigt, 2011, 2020; Voigt et al., 2015).

While TN do not provide a causal graph, we tried to develop one based on the authors’ argumentation (Figure 1, solid arrows). Accordingly, TN (2016a) conceptualize length as a proxy for constitutional restrictions. They argue that length relates positively to both constitutional rigidity and the number/frequency of constitutional amendments. The latter variable, in turn, affects economic success positively. However, controlling for the number of constitutional amendments and for standard macroeconomic variables (e.g., trade openness, education), the authors empirically find a negative impact of length on economic success in OECD countries, operationalized by GDP per capita.

Figure 1.

Causal graph for TN’s theoretical model (2016a) with own additions.

We added two concepts (in red boxes) and several connections (dashed arrows) to the causal graph. One of the concepts is the general disposition of constitutional locking as an unobserved variable. We expect this variable to positively affect both the length and rigidity of a constitution. The second concept is social trust. In societies with high levels of social trust, we expect lower dispositions of constitutional locking and higher levels of economic success.

While we agree that the underlying question concerning the effects of constitutional design on economic performance is important, we will show in our note that the results of TN (2016a) lack empirical robustness. Several minor adjustments to the models, each on its own, reduce or eliminate the presumed relevance of constitutional length.

Replication analysis

Alternative operationalization of a control variable

In our replication analysis, we focus on the results displayed in three tables of TN (2016a: 475; 476; 471, tables 4, 5 and 2) which we were able to reproduce. Here, the authors try to show that constitutional length harms economic performance. In their analyses, they control for several variables, among them the “Number of Amendments under Democracy.” The respective data is drawn from the CCP datasets, which only lists “the years in which the constitution was successfully amended” (TN, 2016a: 462), but not the actual number of revisions. Therefore, the operationalization used for this variable underestimates the number of amendments whenever there was more than one amendment per year. A more important point concerning this variable is the fact that the authors do not explicitly control for the time the constitution is in force although they argue that the probability of successful amendments is related to shifting constitutional cores which occur over time (TN, 2016a: 466). Other things being equal, it seems plausible to expect a constitution to undergo more amendments the older it is.

Therefore, it would be more appropriate to use “Amendment Frequency,” that is, the number of amendments events divided “by either the age of the constitution or the age of democracy, whichever was more recent” (TN, 2016b: 1), a variable which is available in the authors’ dataset. While both measures correlate highly (r = 0.49, p < .01), we find several observations where the respective values diverge considerably (see Figure A.4 in the appendix). Mexico, for example, shows a small number of amendments while its amendment frequency is the highest in the sample. Conversely, Norway reveals the highest number of amendments, whereas its amendment frequency yields a medium rank. Even though amendment frequency still underestimates the true number of amendments as it only counts the number of years with amendments, it at least takes into account the age of democracy/constitution. Thus, it provides a more realistic assumption of the relative frequency of amendments.

As a result, the effect of constitutional length diminishes considerably for the models displayed by TN in their table 4 (2016a: 475; cf. our Figure (A)1 and Table A.1 in the appendix). Its coefficients show only about half the size in our revised models (1a to 6a in Table A.1) compared to the original version (models 1 to 6 in Table A.1). For amendment frequency, we receive much larger standard errors than TN (2016a) do for their operationalization (number of amendments under democracy). The sensitivity analysis (following Cinelli and Hazlet, 2020) at the end of table A1 illustrates that the robustness of constitutional length against unobserved confounders decreases considerably if we use the adapted operationalization for amendments. Our results for constitutional length remain robust if we drop this control variable altogether or if we use the TN’s operationalization and control for age of democracy in addition (not displayed). The only exception to this is model 5 where the TN’s coefficient for constitutional length hardly changes under these modifications.

The effects of our adjustment are even more pronounced for TN’s models in their table 5 (TN, 2016a: 476). Here, “Detail” is the central variable instead of constitutional length, defined as the length of the constitution divided by the number of topics (TN, 2016a: 468). In our modified models using amendment frequency instead of the number of amendments (Figure A.2 and Table A.2 in the appendix, models 1a to 6a), the coefficients for detail lose between 40 and 87% of their size compared to TN’s results. Again, the robustness values of detail displayed in the sensitivity analysis at the end of Table A.2 decrease clearly. If we forgo the control variable for amendments, detail loses even more of its coefficients’ sizes (80 to 91%, not displayed). However, if we keep TN’s operationalization for constitutional amendments and control for age of democracy additionally, the results for detail stay more or less the same as in TN’s original version (not displayed).

Alternative dependent variable

In their analysis, TN use log GDP per capita for the period between 2006 and 2011 as the operationalization for their dependent variable, economic success. While this is a standard operationalization for the concept, it may be misleading if it is considered in isolation. It is plausible to expect GDP per capita to depend on its starting level at the time of democratization or the enactment of the current constitution. Thus, the average annual growth rate of GDP might provide additional hints about the relationship between constitutional length and economic success. The data for this alternative dependent variable stem from the Maddison Project Database (2020). We consider the whole period starting from the democratization or the enactment of the current constitution for the respective countries.

With this modification, we observe extremely large standard errors for both measurements of constitutional length (models 1b to 6b in Tables A1 and A2 in the appendix). In half of the models, the signs for the constitutional length and detail even turn positive. The robustness values for “Detail” displayed in the sensitivity analysis at the end of Tables A.1 and A.2 are (with few exceptions) even lower than the ones for the modification described above (alternative operationalization of the control variable for amendment frequency).

Period of investigation

For their control variable of the number of amendments, TN consider the whole period from democratization or the enactment of the current constitution for the respective countries. Thus, the period of investigation differs strongly. For the US or Belgium, it starts in 1789 and 1831, respectively. For several other countries, such as Finland, Poland, or Switzerland, the period under consideration only starts in the late 1990s. Therefore, we checked if TN’s results hold if we restrict the analysis to the period from 1990 onwards (models 1c to 6c in Tables A.1 and A.2).

In this case, the coefficients for constitutional length (Table A.1) and detail (A.2) remain negative and are sometimes even larger than the ones of TN. However, the standard errors rise even more. As a result, we only see one model (5c in Table A.1) where the effect of constitutional length on economic success reaches a significant level. The values of the sensitivity analysis are once more considerably lower than the ones of TN’s original models, hinting to a decreased robustness of constitutional length/detail against unobserved confounders.

Role of social trust

A last point concerns the effect of social trust on economic performance. TN’s discussion of already existing papers that analyze the relationship between constitutional length, social trust, and economic performance remains somewhat fruitless. Although the authors cite the studies of both Montenegro (1995) and Bjørnskov and Voigt (2014), they leave their findings untapped. Montenegro (1995) perceives constitutional length as a proxy for distrust in a given society: The longer the constitution, the higher the level of distrust. In fact, Montenegro finds that long constitutions are associated with low levels of GDP per capita and that no country with a high GDP per capita has a long constitution. Bjørnskov and Voigt (2014: 91) examine the relationship between “constitutional verbosity and social trust” and find that national trust levels “indeed are robustly and negatively associated with the length of countries’ constitutions”. Both studies thus stress the importance of social trust. Moreover, more recent contributions emphasize the relevance of cultural traits as well. Tarabar and Young (2021), for example, find a positive relationship between cultural individualism and constitutional amendment rates. Ahmad and Hall (2017: 4) refer to numerous studies “finding robust positive impact of social capital, measured by trust variable obtained from the WVS or other comparable surveys, on economic growth”. However, TN refrain from controlling for social trust or other cultural traits. They argue: “cultural arguments are poorly suited to addressing the questions we raise below because they risk overlooking the interactions between culture and constitutions. In other words, it is entirely possible that constitutions influence cultures rather than vice versa” (Tsebelis and Nardi, 2016a: 459). Against the background, it is notable that TN omit an analysis of the possible interactions between constitutional and cultural traits. Thus, they risk causal confounding (Voigt, 2011: 322; Callais and Young, 2021: FN 7).

We reran TN’s models of their table 2 (TN, 2016a: 471) by simply adding generalized social trust (based on data from the World and European Value Surveys) as a control. In all six models, social trust is the only relevant variable (significant at least on the 5%-level). While the sign for constitutional length remains negative, its coefficients lose between a third and a half of their size (cf. our Table A.3 and Figure A.3 in the appendix). The same is true for the robustness values of constitutional length shown in the sensitivity analysis at the end of Table A.3.

Conclusion

To sum up, we can state that the statistical results in the important and pioneering contribution by TN (2016a) are not robust as far as the effect of constitutional length on economic performance is concerned. In our contribution, we tested the effect of several adjustments to the models presented by TN. Each of them substantially reduces the explanatory power of TN’s central independent variable.

Apart from our critical remarks regarding the statistical analysis, we doubt that the focus on constitutional length is a promising path to follow for several reasons. Firstly, constitutional rules differ regarding how binding they are. For example, some constitutions distinguish fundamental rights (e.g., freedom of speech) from aspirational goals (e.g., right to work, right to housing). While fundamental rights are judicially enforceable, aspirational goals are not (TN, 2016a: 466). Secondly, the differentiation between “de jure” and “de facto” rules also needs to be taken into account (Voigt, 2020). This is not only true for authoritarian states that often do not act according to their constitutional regulations, but also for democratic entities. The constitution of the German Land Hesse, dating back to 1946, for example, contains regulations that have the potential to influence the economy heavily. Among these are the prohibition of lockouts (Art. 29 sec. 5), the possibility to prescribe measures necessary to manage the production, manufacture and distribution of goods (Art. 38 sec. 1), and the socialization of key economic sectors (Art. 41). However, none of these provisions has ever been applied. On the other hand, provisions of constitutional rank are not necessarily part of the constitution. If there is a judicial review, courts usually interpret, refine, and develop the constitution. For example, the Swiss constitution of 1874 contained only a few basic rights. In the ensuing decades, the Federal Supreme Court's rulings led to an expansion of this area. The resulting new rights were only formally incorporated into the constitution within the framework of the total revision in 1999, though. Thus, for many years, the text of the Swiss constitution did not reflect the status of several basic rights. As the mere number of words cannot account for this, this measure is just too simple.

Although our paper highlights some weaknesses, there is room for further improvement in future research on this important question of whether constitutions have an impact on economic performance. For instance, neither the intensity nor the importance (i.e., the substantive content) of constitutional amendments has been considered so far. Accordingly, future studies could evaluate their effects.

Supplemental Material

Supplemental Material - Do long constitutions really Hamper economic performance? A comment on Tsebelis and Nardi (2016a)

Supplemental Material for Do long constitutions really Hamper economic performance? A comment on Tsebelis and Nardi (2016a) by Martina Flick Witzig and Adrian Vatter in Research & Politics