Constitutional verbosity and social trust

We are definitely not the first to be interested in the lengths of constitutions. There is, e.g., a small literature asking whether U.S. states with a more independent judiciary are less likely to amend their constitutions—and, in all likelihood, increase their lenghts—because in those states the “cheaper” option of passing new legislation would be sufficiently credible to convince those who demand special-interest legislation of the likely endurance of the newly passed legislation. Crain and Tollison (1979), Anderson et al. (1989, 1990) are examples of this literature.

Reksulak et al. (2004) inquire into the efficiency of language development from a more general point of view. They find that language as a spontaneous order does, indeed, display characteristics of a network good. In particular, they find that the number of additional words of the English language—as documented in the Oxford English Dictionary—is declining with increases in wealth and population.

This is similar, but not identical to the insurance function of constitutions mentioned by Ginsburg (2002). There, politicians who cannot be sure to constitute the majority tomorrow agree on introducing encompassing judicial review to establish a non-majoritarian actor who, they hope, will interpret the constitution according to the intentions of the framers. There are a host of questions closely related to the attempt of current societies to try to bind future societies via constitutional detail. We merely mention some of them here without discussing them, however: what is the relationship between constitutional detail (measured in the length of a constitution) and the formal difficulty of amending it? Are very precisely worded constitutions less likely to be implemented in practice than more open ones? On what normative grounds can current societies ever aspire to constrain the ways future societies make choices and so on.

Another area in which values and norms are extremely stable over very long periods of time has just been described by Alesina et al. (2013). They report a stable relationship between agricultural societies in which the plough traditionally played a dominant role and unequal gender norms today. Persistence holds even among immigrants into European countries and the United States although all of them were born and raised in the ‘new’ country.

Our major source for this information is the Encyclopaedia Britannica (2013). However, in many cases, we employed Wikipedia as an alternative source of information. We note that the two sources virtually never disagree.

In the full dataset, the observations for 110 countries are almost complete. Note that as Luxembourg is not covered in any databases of historical GDP data, the country drops out when adding GDP.

In principle, constitutional length and detail are count data, and we would therefore want to use a Poison estimator or similar discrete data estimator. However, the data in practice form 110 categories and do not form clusters of values, but are distributed as a continuous variable. In such cases, OLS remains consistent and is more efficient than the count data alternatives.

Among other instrumental variables used in previous versions are the severity of the winter climate (temperature average and variability), whether or not the predominant language allows dropping the personal pronoun, the level or salience of religious beliefs, and dummies for whether the predominant religion is Protestant, Catholic, Muslim or Eastern (Hindu or Buddhist). Combinations of these variables also yield good test statistics, as first stage F statistics are above 10 and no combinations exhibit significant Hansen J statistics.

Note that we also tested for the potential influence of predominant religions. When entering regional controls in the specification, we find no significant evidence of religious differences (F=.58; p<.719). Similar non-results apply to a long list of variables, including ethnic, linguistic and religious diversity, geography (country size, latitude and island status) and whether or not countries are former colonies.

As such, when Kurrild-Klitgaard (2001) finds that current levels of economic development are negatively associated with the lengths of constitutions, it therefore seems unlikely to be the result of wealth-based selection of constitutional features. Instead, the combination of a negative association between current wealth and constitutional garrulity and no association between garrulity and wealth levels when the constitution was written suggests either of two possibilities: (1) that shorter constitutions are somehow conducive to very long-run growth; or (2) that certain factors are common to long-run growth and the ability to write relatively short constitutions.

The point estimates (standard errors) when the other two instruments are entered individually are: Flag ratings −101.978 (75.875) and the Nordic dummy 237.987 (4931.473). None of the variables are redundant as instruments, with test statistics of p<.01 (Nordic), p<.056 (monarchy) and p<.01 (flag ratings). Finally, as shown by Bjørnskov and Méon (in press), this particular set of instruments is likely to provide identification of trust across the sample, thus avoiding the problem of heterogeneous first-stage identification.

In a working paper version, we provide simple evidence suggesting that constitutions that are overlong, by the arguments set out in Sect. 2, are indeed associated with better long-run economic consequences. To keep the argument succinct and the paper short, we do not present this evidence in the present paper.