Growth in Autonomous Cities

A common interpretation of how autonomous cities emerged in Europe is that they began as acts of usurpation of authority by groups located in a specific place engaging in a specific type of activity who sought to manage their own affairs rather than having them be managed by a feudal ruler. Max Weber referred to this more specifically as a regime of “non-legitimate domination.”Footnote ¹² If many merchant dominated cities were initially based on informal, oath-based associations, over time the need to establish property rights amongst a larger number of individuals led to the creation of more formal institutions of rule, and in the early stages of their development, that is, up until the fourteenth century, these formal rules very often ensured that a city's merchants guild held all the reins of political power. Cities most generally had a large city council with broad membership and then a smaller governing council composed of a group of magistrates. In a number of cities it was specifically stipulated that only members of a city's merchants guild or like individuals could hold one of these positions. As an example, in Siena under the rule of The Nine, a statute in 1287 stipulated that members of The Nine must be of the merchants of the city or similar individuals.Footnote ¹³ Merchant guild control was further favored by the fact that a large number of cities, particularly in Northern Europe, chose their magistrates by a system of cooptation. Rather than having magistrates be elected by the citizenry, in a system of cooptation an outgoing body of magistrates would choose an incoming group.Footnote ¹⁴ The data reported in Stasavage (Reference Stasavage2011, 58) for a small sample of autonomous cities show that prior to 1300, merchants held an average of 72 percent of the seats on governing councils. The key characteristic of an autonomous city was not simply that autonomous cities had merchants guilds, it was that the guilds also had a very direct degree of political control. The fact that guilds ultimately depended on political support for their policies has been heavily emphasized by Epstein (Reference Epstein2000; Reference Epstein, Epstein and Prak2008).

The pattern of governance in many autonomous cities changed during the fourteenth century, although not in a way that fundamentally altered the core characteristics relevant for this article. Autonomous cities continued to provide property rights for citizens with barriers to entry to outsiders. Autonomies cities continued to be governed by guilds that provided property rights for insiders to the exclusion of others. The change was that in a number of cities merchants guilds were now forced to share power with members of the craft guilds. During the course of the fourteenth century, and in some cases into the fifteenth, a large number of Europe's autonomous cities experienced popular revolts in which members of craft guilds engaged in light industrial production challenged the dominance of merchant guilds. Some of the best known of these include the revolt that took place in Ghent in 1302 and the Revolt of the Ciompi in 1378 in Florence. In some cases, such as in Ghent, these revolts led to durable changes in institutions for governance. In other cases, as in Florence, a prior pattern of rule was quickly re-established. Durable changes in institutions sometimes involved a shift to electing magistrates rather than selecting them via cooptation. They also included opening up seats on governing councils to include guild representation. In these cases governing councils would have a certain number of seats set aside for the merchants guild and the craft guilds. With the exception of those revolts, such as that of the ciompi, when even the lowest orders of society demanded and briefly received representation, expansion of city councils simply extended political power to guild organizations that, like the merchants guilds, provided property rights for insiders but barriers to entry to those on the outside. Finally, in the aggregate the extent to which the political systems of city states were truly reformed is open to question. According to the data reported in Stasavage (Reference Stasavage2011, 58) by the beginning of the sixteenth century merchants still held nearly 60 percent of the seats on governing councils.

There is good reason to believe that the form of autonomous city rule referred to above could bring economic benefits, at least in the short run. Governance of a city council by merchants and consequent insulation from the whim of an outside monarch may have made for a more stable legal environment in which to conduct business (Mokyr Reference Mokyr1995; Reference Mokyr1990). In addition, governance by a group of individuals meeting regularly face to face may have facilitated the sort of commitment mechanisms described by Avner Greif (Reference Greif2006). Finally, recent historical work has emphasized that guilds were not simply rent-seeking bodies. They played a very prominent role in property rights provision, in particular the protection of knowledge, and in the facilitation of training.Footnote ¹⁵ These political conditions may have initially provided an environment both for increased trade and increased innovation. Therefore, in the period after establishing its autonomy we might expect an autonomous city to grow more quickly than before.

However, precisely because the establishment of an autonomous city involved the usurpation of authority, it also involved the creation of barriers to entry as the merchant, and later, craft guilds that dominated a city's political institutions strictly regulated commerce and the right to enter certain professions. Henri Pirenne described this particular view of the economic effects of guild control in the following terms:

This negative view of the combined effect of guild control with political control has of course been emphasized by numerous other authors already cited in the introduction to this article.Footnote ¹⁶

The above discussion points to advantages and shortcomings of the political regimes of autonomous cities. Rather than attempt to adjudicate whether it was the advantages or shortcomings that dominated in a static sense, a more novel way to think about things is to consider how these effects might vary over time. In fact, there are reasons to believe that in an initial phase of development, the advantages might dominate but then in later stages the shortcomings would play a more preponderant role.

One reason we might expect an autonomous city to first prosper and then stagnate is if we refer to the work of Mancur Olson (Reference Olson1982) and suggest that over time economic stagnation would have been produced by a progressive accretion of rent seeking vested interests. But it appears that in most cases, vested interests and barriers to entry were an important feature of Europe's autonomous cities right from the start. This is inconsistent with an Olsonian account in which such vested interests and barriers to entry emerge gradually over time.

A more convincing reason why we might expect autonomous cities to initially prosper and subsequently stagnate can be derived from the work of Daron Acemoglu (Reference Acemoglu2008). Acemoglu suggests that oligarchies in which members of the oligarchy are themselves entrepreneurs may initially have high rates of growth because of better protection of property rights and a lower risk of expropriation. However, if it is necessary to have “churning” in the identity of entrepreneurs to maintain a rate of economic growth, then oligarchies will eventually stagnate. He suggests that this could be the case if the entrepreneurial skill of an individual, or of a family dynasty, changes over time, necessitating new entrants. It could also be if the entrepreneurial skill of an individual or dynasty is constant over time but comparative advantage in entrepreneurship changes over time as the type of economic activities in an economy evolves. This phenomenon could also be related to the well-known “replacement effect” proposed by Arrow (Reference Arrow1962) whereby a potential entrant to a market has a greater incentive to innovate because their alternative to innovation is zero profit whereas an existing firm could continue to enjoy a monopoly stream of profits.Footnote ¹⁷ Any of the above possibilities could clearly apply to the economies of medieval and early modern cities.

Autonomous and Nonautonomous Cities Compared

So far I have said little about cities under the direct control of territorial princes, as opposed to those that were politically autonomous. It is important to emphasize again that autonomous cities were not alone in having guilds. The key difference I have argued was that in autonomous cities guilds had political control whereas guilds in cities under princely domination were subject to an outside force. This then raises the question of what incentives territorial princes faced with regard to their cities.

To consider the above question in an abstract fashion, take the case of a territorial prince who is not directly engaged in economic production and who has a desire to extract revenue from a city. Under princely domination, the prince will set a tax rate that maximizes revenue subject to the constraint that a city's merchant and/or craft guilds might produce less or trade less. Under urban autonomy, as long as the guilds that control the city's government derive less benefit from princely revenues than does the prince, then they will set a lower tax rate and economic production will therefore be higher under autonomous rule. If this was all that mattered we would therefore expect growth to be higher in the case of urban autonomy. However, once we introduce the possibility of barriers to entry into the city's markets, then we see that this might not necessarily be the case. If a city is autonomous then a guild or guilds that control its government will have an incentive to establish barriers to entry because the monopoly power that they provide leads to increased profits with the accompanying outcome of lower total production, the classic result for monopoly. Since lower total production implies lower taxable revenue, under princely domination a prince should face incentives to oppose the imposition of barriers to entry into a city's markets. If this was all that mattered, it would imply higher production and higher growth under princely domination.

We can quickly see from the above discussion that without further knowledge about the values of the parameters determining how high a city's guilds would set barriers to entry and how high a prince would set taxation, we can make no clear prediction about whether economic production will be higher in an autonomous city as opposed to one subject to princely control. However, we can nonetheless use the above theoretical sketch to make a prediction about the time path for growth in an autonomous city relative to a nonautonomous one. If the effect of barriers to entry becomes more pernicious over time, precisely because they stifle innovation, then we should expect any growth advantage of autonomous cities to decline over time.

To what extent does the above discussion bear any resemblance to the actual behavior of European territorial princes? That princes imposed taxes on cities is hardly in doubt. But what about the idea that princes might actually insist that cities maintain lower barriers to entry into markets? This is an idea that has been emphasized by Epstein (Reference Epstein2000, 101–105, 125) and which has also received qualified support from Blockmans (Reference Blockmans2010, 585–586).

In his book Freedom and Growth, Stephan Epstein (Reference Epstein2000) made use of a comparison between the evolution of the textile industry in Lombardy and Tuscany during the fourteenth century. While both of these regions during the twelfth and thirteenth centuries consisted of numerous autonomous cities, during the first decades of the fourteenth century Lombardy was transformed into a territorial principality ruled by the Visconti family in which city autonomy was essentially abolished. A similar fate would eventually befall most of Tuscany's independent cities, although not until considerably later under the domination of the Medici. If both Lombardy and Tuscany had craft guilds during this period, in Tuscany the guilds exercised direct political control through the councils of autonomous cities that controlled their own economic policy, whereas in Lombardy this was not the case. Under the Visconti, as early as 1346 Lombard towns lost the right to create such restrictions. Epstein suggests that this abolition of barriers to entry resulted in greater innovation and production in Lombardy relative to Tuscany.

We see a similar effect of princely control in Flanders.Footnote ¹⁸ Flanders was dominated by three principal cities, Bruges, Ghent, and Ypres, each of which had a substantial textile industry and where guild involvement in politics resulted in the establishment of very significant barriers to entry for new artisans. However, political autonomy for these three principal cities was not as complete as in the case of the Tuscan communes of the fourteenth century. Each city was still subject to at least nominal overlordship by the Count of Flanders. One consequence was that the effective degree of autonomy often varied depending on who exactly was count. During the initial decades of the fourteenth century the three great cities enjoyed very substantial autonomy and they used this autonomy not only to maintain barriers to entry but also to ensure that no competing producers established themselves in the smaller surrounding towns and countryside. However, after 1349 under a new Count of Flanders there was a very substantial shift in power back to the Count's institutions and away from the guilds in the autonomous cities. One consequence of this reassertion of princely authority was that Bruges, Ghent, and Ypres no longer found it possible to exercise influence over the development of the textile industry in the surrounding towns and countryside. Once again, princely control over cities led to lower barriers to entry.

There is also a further element one might consider with regard to both autonomous and nonautonomous cities, although this will not be the main focus of my empirical tests. The ability of cities to establish barriers to entry may have depended not only on whether they were autonomous, but also on whether they had representation in national level representative institutions. While many national representative assemblies did not have prerogatives allowing them to heavily influence commerce, others were most certainly engaged in such activity. The assemblies in Flanders that have been extensively studied by Blockmans (Reference Blockmans1976) are a good example here.

To return to my main point, the above discussion provides both a theoretical and a historical basis for my conjecture that relative to nonautonomous cities, politically autonomous cities should first experience a high rate of growth followed by subsequent stagnation. The next step is to assess this possibility empirically, beginning with a presentation of my autonomous city data set.

Placebo Test for Preautonomy Growth

I consider the possibility that rapid growth for autonomous cities reflected an underlying trend by conducting a variant of a placebo test. As discussed above, this test is similar in spirit to the Granger test for difference in differences models proposed by Angrist and Pischke (Reference Angrist and Pischke2009, 237), and it is designed to consider the possibility that my results are biased by the failure to take account of the possibility that growth caused cities to become autonomous in the first place.Footnote ³² Taking the specifications in Table 6, I recoded the three variables A, Y, and Y ² by setting a placebo date for the establishment of autonomy equivalent to 100 years prior to the actual date. I then re-estimated each of the six specifications using these recoded variables. As can be seen in Table 7, the results are fairly unambiguous. The coefficients on A are smaller than in the Table 6 estimates, and they are generally not statistically significant. The same conclusion applies for the linear trend terms. This is strong evidence that the fact autonomous cities grew more quickly than nonautonomous cities does not reflect an underlying trend that commenced prior to the establishment of autonomy. The possible implication then is that cities were able to establish their political autonomy not because of a strong period of prior growth, but due to factors such as geographic isolation from the capitals of princely rulers. As I have noted above, there is robust evidence that the manner in which the Carolingian Empire fragmented made it easier for some cities to subsequently establish their autonomy. This interpretation is also supported by the tests in Tables 2 and 3.

TABLE 7.

Placebo Test Using OLS Estimates for City Population Growth

Notes: Dependent variable in all specifications is the percentage change in city population. N = 1052. “Controls for observables” includes dummy variables for oceanic ports, riverine ports, bishop's seat, roman settlement, in addition to latitude, longitude, and the product of latitude and longitude. The coefficients for these variables are not reported. The specifications including “(Coordinates)*(Time dummies)” include latitude, longitude, and the product of latitude and longitude, with each of these variables interacted with a full set of time period dummies. The specifications including “(Region)*(Time dummies)” include a set of dummy variables for region (modern-day country) interacted with time period dummies. All standard errors are clustered by city.

A Varying Autonomy Effect by Region

In the estimates I have reported so far, I have implicitly assumed that the effect of political autonomy does not vary by region. The result is that I may have a good estimate of the average effect of autonomy across regions. However, there may be enough variation from region to region in this effect to imply that the average is not particularly meaningful. The effect of political autonomy could vary if the nature of the relationship between princes and rulers of autonomous towns (or other towns) differs from region to region. To investigate this possibility I first re-estimated specification 3 from Table 5 while interacting A with a set of region dummies where region corresponds to a modern-day country. In specification 3, which includes city fixed effects, the coefficients of these interaction terms were relatively precisely estimated. Even so, a test showed that one could not reject the null that the coefficients on these interaction terms were identical. A very similar result was obtained when either the A or Y variables were interacted with a set of region dummies using specification 3 in Table 6. Once again the interaction terms were relatively precisely estimated, but it was not possible to reject the null that they were identical.

The above test suggests that the effect of urban autonomy did not vary tremendously by region. There are other ways that one might approach the problem. We could posit a more explicit theory about the factors influencing autonomy and then attempt to test this proposition more directly. One possibility is that the effect of urban autonomy depended on country size. The opportunity cost of barriers to entry in autonomous cities may have been greater in larger markets. To the extent that polity size proxies for market size we would then expect the effect of autonomy to be smaller in larger countries. However, we know that many of the large countries in Europe at this time had very fragmented markets due to restrictions at the regional level, and France provides an excellent example of this. All the same, I attempted a crude test of this hypothesis by breaking my sample into cities located within regions where political jurisdictions were large (France, England, Castilian Spain, Southern Italy) and where political jurisdictions were small (Austria, Germany, Northern Italy, Portugal, and Switzerland). I then included an interaction term where my autonomy variable A was interacted with political jurisdiction size. I then repeated my core specifications from Tables 5 and 6. In no case was this interaction term statistically significant.

A second possibility, and one that I have already referred to above, is that the effect of urban autonomy may have depended on the extent to which cities (autonomous or not) had power within national representative institutions in those cases where such institutions existed. Power within national representative institutions may have helped allow cities (autonomous or not) to establish barriers to entry. One could potentially use either the database of representative institutions produced by Stasavage (Reference Stasavage2011) or that produced by Bosker, Buringh, and van Zanden (Reference Bosker, Buringh and van Zanden2013) to investigate this possibility. However, in order to do so we face a thorny problem in that some of the autonomous cities in my sample can be said to have been located within a larger polity whereas others were more completely independent. An effective test of this hypothesis would therefore first depend on deciding which ones to include in the analysis. A second problem is that existing representative institutions data may be poorly suited to considering the question. The database constructed by Stasavage (Reference Stasavage2011) focuses on assembly prerogatives in the area of public finance, but not the commercial questions that would be most relevant for the core question of the current article. Bosker, Buringh, and van Zanden (Reference Bosker, Buringh and van Zanden2013) have produced a measure showing the frequency with which parliaments that included urban representation met. However, their data contain no information on prerogatives with regard to commercial regulation.

A Varying Autonomy Effect by Time

We might also want to consider whether my econometric results are potentially biased by a failure to control for a changing effect of autonomy across historical time periods. This is made difficult by the fact that cities became autonomous during a fairly narrow time window during the twelfth and thirteenth centuries. My inclusion of period dummies in all of my specifications controls for the most obvious source of bias. We know that the thirteenth century was a period of fast growth in Europe relative to what would follow, and we also know that there were numerous newly autonomous cities at this point. The inclusion of location controls interacted with time in several of my specifications controls for a further potential source of bias. There were certainly time and region specific shocks to growth in Europe during this period, and it may be the case that these shocks were correlated with the fraction of cities that were autonomous in each region as well as the length of time that they had been autonomous. The final remaining possibility is that there may have been shocks that affected autonomous and nonautonomous cities asymmetrically. Perhaps it was the case that autonomous cities were good at delivering growth during a particular historical period and not afterwards.

Following on the above intuition, received wisdom from numerous authors suggests that some time around AD 1500, a series of technological changes, and in particular the introduction of the cannon, led to increased fixed costs for defence because cities now had to build sturdier walls than ever before.Footnote ³³ This may have necessitated higher tax rates with the knock on effect of lower growth. If true, this would provide an alternative account to mine for the time path of autonomous city growth. To test this possibility I estimated the following equation:

(2) \begin{eqnarray} g_{it}&=&\alpha +\beta A_{it}+{\gamma }_{1}Y+{\gamma } _{2}Y^{2}+\lambda (A_{it}\cdot {(t\ge Y}^{\ast }{)})+\zeta\,\, p_{it}\nonumber\\ &&+\,\mu _{i}+\theta _{t}+\varepsilon _{it}. \end{eqnarray}

This equation is the same as that used for the Table 6 specifications, augmented by a parameter λthat estimates the changing effect of city autonomy after a prespecific break date Y* . I consider three possible break dates: 1400, 1500, and 1600. If the military technology interpretation is accurate, then we should expect λ to be negative and statistically significant in at least one of the specifications. Likewise, if my existing regression results have been biased by failing to take account of this shift in military technology, then my estimates of ${\bm \gamma}_{1}$ and ${\bm \gamma}_{2}$ should be closer to 0. In Table A4 in the Appendix I report results where I re-estimate the first three of the Table 6 specifications while adding the term λ(A_it · (t ⩾ Y*)). This term is simply A multiplied by a dummy variable that takes a value of 1 for all years beginning in year Y* and 0 otherwise. The results are as unambiguous. While my estimates of ${\bm \gamma}_{1}$ , and ${\bm \gamma}_{2}$ remain very similar, the estimate of λ is generally not statistically significant, and when it is statistically significant the coefficient is positive, rather than being negative as the military technology interpretation would suggest.

The test proposed above suggests that my findings regarding the time path for growth of autonomous cities is not produced by the fact that autonomous cities suddenly began to fare worse as a result of military changes that took place between 1400 and 1600. As a next step, we might also want to consider whether more gradual pan-European changes produced the reversal in fortune between autonomous and nonautonomous cities. In order to explore this possibility I augmented the Table 6 specifications to estimate the following equation:

(3) \begin{eqnarray} g_{it}&=&\alpha +\beta A_{it}+{\gamma }_{1}Y+{\gamma } _{2}Y^{2}+\lambda _{1}A_{it}t+\lambda _{2}A_{it}t^{2}+\zeta\,\, p_{it}\nonumber\\ &&+\,\mu _{i}+\theta _{t}+\varepsilon _{it}. \end{eqnarray}

The idea here is to use the estimates of the γ and λ parameters to test between two potential trends. To the extent that the γ parameters are statistically significant, we can conclude that the effect of political autonomy depended on how long a city had been autonomous. To the extent that the λ are statistically significant, we can conclude that the effect of political autonomy varied by time period. Given that all autonomous cities emerged during a specific period in time, t and Y will be highly correlated, and this is therefore quite a demanding specification. The results of this exercise are reported in Table A5 of the Online Appendix. They do not lead to an unambiguous conclusion, but they do provide some evidence that my Table 6 results do not depend on the fact that the effect of autonomy varied by time period t and not by years of autonomy Y. In none of the three specifications do the λ coefficients approach statistical significance. In contrast, the ${\bm \gamma}_{1}$ coefficient is of essentially identical magnitude to that in the Table 6 estimates that do not include the λ₁ At and λ₂ At ² terms. However, the ${\bm \gamma}_{1}$ coefficient is somewhat less precisely estimated.

Biases Due to Coding Error

A further possible source of bias in my estimates might involve that due to measurement error in my three autonomy variables A, Y, and Y ². As described above, I developed a coding index that has three levels, the the top of which is attained if there is consistent information from more than one of my sources for the city in question. This is the case for 111 of my 173 cities. Measurement error that simply involves idiosyncratic noise would ordinarily be expected to simply inflate my standard errors, but we cannot of course assume a priori that any measurement error takes this form. It could be the case that data for the low information quality cities is biased in a particular direction. As one way to approach the problem, I re-estimated the Tables 5 and 6 specifications while only retaining the 111 cities that have the highest level for the data coding index. The results of this exercise are reported in Tables A6 and A7 of the Online Appendix. Using the high quality sample of 111 cities, in the specifications distinguishing between autonomy before and after Y = 100 we continue to observe no positive effect of autonomy on average, but there remains consistent evidence of a growth advantage for autonomous cities during their early period of autonomy. When repeating the quadratic trend specifications while using the high quality sample we continue to obtain very similar results.

One might also be worried that any bias in my sample would derive from the high information quality cities. It might be the case that cities experiencing periods of fast growth were more likely to be the subject of attention, there would be more information about them, and thus we would be more likely to code them as being autonomous (remembering that the default in the case of low information is to code as nonautonomous). In practice this appears not to have been the case. The population growth rate is only a very weak predictor of the data quality index.

Spatial Correlation

Growth in groups of cities may tend to exhibit patterns of spatial correlation. In the case of medieval and early modern Europe, it is well known that urbanization initially advanced most quickly in two clusters centered around northern Italy and the Low Countries, and that more generally there was a zone of higher urbanization between these two clusters.Footnote ³⁴ I have already controlled for this to a significant degree by including controls for city fixed effects in addition to controls for location interacted with time. However, failure to take account of any remaining spatial correlation in the residuals of my regression estimates could result in biased estimates of the standard errors. The form of spatial dependence in city growth rates could potentially be quite complex, as there are plausible reasons for growth in a given city to be either positively or negatively correlated with that of its neighbors. To take account of the potential effect of spatial dependence, I tested for spatial correlation of errors using a test appropriate for panel data proposed by Pesaran (Reference Pesaran2004). This is a test based on averages of pairwise correlation coefficients of regression residuals. Importantly, it is also a test that does not require a priori specification of a spatial weighting matrix.Footnote ³⁵ Based on this test, in each of the fixed effects specifications in Tables 5 and 6 the test statistic indicated that in all cases it was not possible to reject the null hypothesis of no spatial correlation.

Other Time Varying Unobservables

As a final step, it is worth considering what remaining time varying unobservables could potentially be biasing my results. One possibility is that it might be the case that when a particularly effective leader or leadership assumes control of a city, they would seek to simultaneously establish autonomy and also take actions favorable to growth. If a current leader or leadership is particularly competent, in expectations a subsequent leadership would be less competent, creating the possibility of mean reversion in population growth rates driven by nothing other than leader turnover. In this case I might find a pattern whereby autonomous cities initially grew quickly and then more slowly but for reasons that have nothing to do with the effect of autonomy itself. One way to rule out the above possibility could be to identify a suitable instrumental variable for city autonomy, but this is not an easy task. As shown in a previous section, proximity with the Meersen partition line established at the end of the Carolingian Empire is a very strong predictor of city autonomy as well as its duration. However, my estimates include multiple endogenous autonomy variables, so the model cannot be identified with a single instrument. Moreover, proximity to the Meersen line is likely to be correlated with multiple factors that might influence economic growth, and so it is very uncertain that the exclusion restriction for this instrument would be satisfied. Both of these problems could potentially be solved by instrumenting with the Meersen distance interacted with some function of time, or alternatively with a set of period dummies. Unfortunately, this instrumenting strategy resulted in very imprecise estimates.

If the possibility of time-varying unobservables as confounders cannot be ruled out by econometric means, there may still be other reasons to believe that they are unlikely to be creating bias in my estimates. The reason for this is that if my estimates show a pattern whereby autonomous cities first grew more quickly than others, with a subsequent decline in this growth advantage, they also show that a little more than a century after establishing autonomy, an autonomous city would actually be expected to grow more slowly than a nonautonomous city. Moreover, we know that among those cities that did become autonomous, a very high fraction remained autonomous for more than a century, so this pattern is actually common in the sample. In the case of mean reversion being explained by leadership turnover, this would then imply that a particularly competent leadership would on average be expected to be followed by a particularly incompetent leadership. One would need to provide a theoretical reason for such an expectation as standard properties of mean reversion could not account for it. More generally, the fact that autonomous cities eventually grew substantially more slowly than nonautonomous cities suggests that only a substantially more restricted set of time-varying unobservables could possibly be producing the observed pattern in the data.