Introduction

Much attention has been paid to the level of physical disorder in urban neighborhoods, that is, the deterioration and denigration of buildings and spaces, and what it can communicate about the local community. For over a century, efforts to map conditions across the urban landscape have consistently noted that elements like poor sanitation, dilapidated housing stock, and graffiti cluster in neighborhoods that also suffer from more serious issues, like elevated crime rates, diminished resident health, and poor youth outcomes (Booth 1903; Cohen et al. 2000; Mayhew 1862; Park and Burgess 1925; Sampson 2012). This has led many theorists to argue that the accumulation of physical disorder indicates that a community is unwilling or unable to monitor and manage the public space (Bursik and Grasmick 1993; Jacobs 1961; Sampson and Raudenbush 1999; Shaw and McKay 1942/1969; Wilson and Kelling 1982). Thus, while these failures of maintenance are relatively minor, they reflect a deeper, more systemic deficiency within the community that can manifest itself in a variety of ways. In this manner, physical disorder is an easily observed symptom of the social and behavioral well-being of the community. One popular version of this thesis, known as “broken windows theory,” has even gone so far as to attribute to disorder a causal role, in which these signals of vulnerability encourage further incivilities and delinquent behavior, while also discouraging others from taking action for the good of the community (Skogan 1992; Taylor 2001; Wilson and Kelling 1982).

The study of the correlates, causes, and consequences of physical disorder has been immensely fruitful through the years, spreading from criminology and sociology to community psychology (e.g., O’Brien and Kauffman 2013; Ross and Jang 2000), developmental psychology (e.g., Kohen et al. 2002), and public health (e.g., Theall et al. 2013; Wen et al. 2006), and inspiring a variety of innovations in public policy and practice (e.g., Kelling and Sousa 2001; Skogan 1992). As a metric, however, the information that it provides is unidirectional, running from the complete absence of disorder to extreme levels of neglect and denigration. Put another way, a broken window tells us about failures in the natural functioning of a community, but little about the ways in which it might be thriving. This is representative of a common issue across the social and behavioral sciences—an overemphasis on pathology in both individuals and communities with little attention to processes and symptoms associated with thriving (Seligman and Csikszentmihalyi 2000). Here we seek to address this by developing and examining a measure that captures patterns of proactive investment in a neighborhood—the very opposite of the neglect and denigration signaled by physical disorder. It is intended to be a complementary measure that might capture the other end of the spectrum of neighborhood maintenance and upkeep, providing a more comprehensive and nuanced view of local behavioral and social dynamics.

The question remains, though, how one might measure proactive investment. Just as physical disorder is reflective of failures of maintenance, one might identify visible artifacts that such behavior would leave behind. For example, some have conducted systematic audits of decorations and other personalizations of lawns and porches, finding that they do correlate negatively with disorder and positively with strong relationships between neighbors (Brown and Werner 1985; O’Brien and Wilson 2011; Werner et al. 1989). Another approach would be to directly measure such investment through the construction work done on a neighborhood’s buildings, reflecting efforts by property owners to invest in the improvement of their properties. City governments keep records of such projects in the form of building permit applications. In recent years, there has been a rapidly growing trend to digitize these and other administrative data, making them readily available for study. The current manuscript uses the record of building permits issued in Boston, MA over a three-year period to develop an ecometric (i.e., a measure of neighborhood-level conditions) of investment. Importantly, administrative data such as these were not made for research purposes, and thus the analysis has three main steps: to isolate information that captures the construct of interest; to establish criteria for measuring it reliably across time, laying the groundwork for longitudinal work; and, finally, to assess its statistical relationship with other key neighborhood indicators, particularly physical disorder and local social dynamics, in order to gain a fuller view of its role in the broader ecology. This process is described in greater detail in the next subsection.

Ecometrics and Big Data

The past decade has seen the rapid proliferation of large-scale, digital administrative data sets, each capturing some aspect of daily life. These range from cell phone calls to social media posts, from credit card swipes to requests for government services. Nowhere is this truer than in cities, where there is the greatest density of people using digital technologies, and local governments and companies have the human and financial capital to invest in cutting-edge systems that compile and utilize such information. These “big data,” as they are often termed, span a broad array of domains and feature unprecedented richness and detail, leading some to argue that they will transform the social and behavioral sciences (e.g., Anderson 2008). In contrast, others are less enthusiastic and even wary of big data, noting in particular that such data are the incidental byproducts of an administrative process, not the intentional products of a research protocol, thus casting doubt on their interpretability, and, in turn, any utility they might have for rigorous science. As with most such debates, the truth likely lies somewhere in between these two perspectives: these data are distinctive in their content and precision, making them a potentially important resource, but much care must be taken when incorporating them into research (Boyd and Crawford 2011; Lazer et al. 2009; O’Brien et al., accepted).

How the data should be handled will depend largely on the manner in which a researcher wants to use them; just as there are different methodological guidelines for collecting different types of measures, there need to be different techniques for evaluating the meaning of administrative data. In the current case, the goal is to capture a specific ecometric, or measurement of a characteristic of the physical or social ecology of a neighborhood. Ecometrics have become increasingly popular in urban research writ large, and in community research more specifically, since the publication of Raudenbush and Sampson’s (1999) rubric for their collection (also see Nickelson et al. 2013 for a review). Drawing off of the long-standing traditions of psychometrics, their goal was to encourage survey and observational protocols that could generate ecometrics that were both robust in their content and reliable in their measurement.

Administrative data, particularly those collected by the government agencies tasked with serving a city’s neighborhoods, offer a novel resource for ecometrics. Many describe discrete events or conditions, each occurring at a specific place and time, providing a direct window into the varied landscape of the city. The methodology forwarded by Raudenbush and Sampson (1999), however, was intended to guide original data collection, and did not extend to the extraction of ecometrics from pre-existing data. Taking note of this gap, O’Brien et al. (accepted) forwarded a new methodology for ecometrics in the age of big data, seeking to address three main issues arising from the use of such data. They illustrated this approach by measuring physical disorder from over 300,000 requests for non-emergency government services received by Boston, MA’s Constituent Relationship Management (CRM) system over a 2.5-year period.

First, the need to identify relevant content arose from one of the purported strengths of such data—they were too rich in information to be immediately useful. From an initial set of 178 different types of requests, the authors isolated 33 reflecting physical disorder. Factor analysis produced two subfactors: private neglect, including housing issues (e.g., pests), the uncivil use of private space (e.g., illegal rooming houses), and complaints about big buildings (i.e., condos); and public denigration, including requests for graffiti removal and issues regarding the improper disposal of trash.

Second, there was concern regarding the validity of measures based on constituent calls. If neighborhoods systematically differed in their tendency to report issues, the resultant measures would be biased. To address this, the authors conducted neighborhood audits to measure the “civic response rate,” which they used to calibrate the call-based measures to better reflect objective conditions. The adjusted measures correlated strongly with other indicators of blight, establishing construct validity.

Third and last, it was necessary to establish criteria for reliability as administrative data do not come with guidelines regarding the spatial or temporal windows at which they are most appropriately measured (Hipp 2007). Testing at multiple time intervals revealed that both private neglect and public denigration could be reliably measured every 6 months for census block groups and every 2 months for tracts.

In sum, O’Brien et al. (accepted) provided an initial demonstration of the opportunity administrative records hold for ecometrics. The database supported a multidimensional measure of physical disorder where neighborhood audits have typically captured a single dimension; it could access conditions that would be difficult to measure through observation, like those that are very rare or occur away from public view; and the authors’ procedures can be replicated longitudinally with relative ease and for minimal cost, since the data are generated continuously by the city’s call centers. This stands in stark contrast to the extensive time and effort required for whole-city surveys and observational audits. Further, it provides the basis for similar work with other data sets, potentially leading to a wide array of ecometrics. Some, like private neglect and public denigration, will be modified versions of classical measures, while others, like the one proposed here, will be less familiar.

The Current Study

The goal of the current study is to develop an ecometric of proactive investment by community members using a database of building permits from Boston, MA from 2009 to 2012. The analysis will proceed as follows, adhering closely to the three steps articulated by O’Brien et al. (accepted) in their measurement of physical disorder from 311 calls. The one main difference is that we assess reliability before validity, as is more typical in traditional methodologies. The 311 calls required this distinctive order because the validity tests gave rise to final, adjusted measures that were then the basis of the reliability analysis. We will not be creating composite measures here, as we explain below, making it possible to assess reliability first.

  1. 1.

    Content: It is necessary to first isolate information that reflects proactive investment by community members. Not all building permits will be for projects undertaken by homeowners or local landlords. In some areas of the city, there are ongoing, large-scale developments being undertaken by large firms and corporations, and it is important to distinguish these from the more organic process of community-based investment.

  2. 2.

    Reliability: The database covers multiple years and is relatively large (150,493 permits in ~2.5 years), suggesting that it might be divided into multiple time windows, each capturing investment across neighborhoods for a more precise period. The primary intent of this analysis will be to identify the optimal time window, defined as the smallest one that is still robust against measurement error. This also sets the stage for longitudinal analysis and the potential ability to examine long-term trajectories.

  3. 3.

    Validity: The concerns of validity here are somewhat different from those from a call record, like the 311 data. Building permits are not subject to the whims of constituent calls but are objective instances in which an individual requested permission to do work on a building. One advantage of these data is that they should be robust against “double counting,” as a project at a specific address can be discretely identified despite multiple permits. There does remain the possibility of false negatives, as some building owners may undertake projects without applying for the necessary permits. Because permits do cost money (~$50 on average), it might be assumed that such undercounting will be more common in lower income areas. This can then be controlled for in any analysis by including median income as a correlate. In any case, it is worth noting that these will be the smallest projects, thus limiting the overall loss of information.

    Because this measure is meant to fill a gap in current methodologies, it is not clear how one would validate it against a neighborhood audit. Instead, we strive here for construct validity, assessing the statistical relationships that our measure holds with other popular neighborhood measures in order to better understand it as a component of the local ecology (Messick 1995). It is useful then to have already assessed reliability as we will want to conduct correlations using time windows that are appropriate for measurement. This analysis pays particular attention to three details. First, we examine correlations with those things that logically should be associated with investment, most importantly homeownership and access to resources (i.e., median income). Second, we attend to one of our fundamental hypotheses, that this measure complements measures of physical disorder, comprising the other end of a spectrum of neighborhood maintenance. Using regressions we test whether the correlation between these items is a result of shared demographics, or whether the two are connected behaviorally in some way that cannot be accounted for by mere demographics. We conduct this analysis using the measures of physical disorder provided by the 311 system. We also utilize a survey measure of collective efficacy, or the ability of neighbors to leverage local relationships and social norms in the management of the neighborhood (Sampson et al. 1997), in order to see the relationship between investment and these critical social processes.

    Last, it is important to note that our measure might not exclusively indicate thriving within a community, but might also reflect gentrification, in which the space might be improved, but not necessarily by the people who have been living there. We also conduct a preliminary, visual analysis of this question, mapping investment across the city and identifying regions where it is elevated when considering local socioeconomic characteristics.

Methods

Data Sources and Measures

Building Permits Database

As of December 15, 2009, the Inspectional Service Department of the City of Boston began storing all information pertaining to applications for building permits in a digital database. The permits database analyzed here begins on that date and runs through June 26, 2012, including 150,493 building permit applications. Each record describes a single permit application, including: the address at which the work would occur; the type of permit requested, reflecting the specific nature of the work to be completed; a more detailed description of the work (a subtype of the permit); and the date on which the permit was issued. The city issues 55 different permit types, all of which were represented in the database. Analysis here is limited to those permits that could be mapped to a known address (n = 142,592). All permits in the database had been approved, but 26,573 had no issue date entered, in many cases because the permit in question is not actually “issued” (e.g., appearing before the board of appeals). For these, the date of application was also used as the date of issue. For those permits of types that are issued, issue dates were imputed using the average time between application and approval for other permits of the same type.

Because a single project might require more than one permit, we use parcels (the smallest unit to which a deed might be held; all addresses have at least one parcel, and some have more (e.g., condo buildings)) as the primary unit of analysis. All permits for a single parcel were combined to create parcel-level measures (e.g., the number of permits requested) that might best capture the work being done there, as described in greater detail in the results. Another advantage of this approach is the suppression of “false positives,” or double-counting of permits, leading to overestimates of investment in certain areas. Instead, if multiple permit applications were generated by a single parcel their composite can be treated as a single project. Of Boston’s 336,935 parcels, 45,092 requested at least one permit. Using a broader database of the places and regions of Boston, all parcels were attributed to the appropriate census geographies (from the 2005 to 2009 American Community Survey, the most recent census with socioeconomic data when the database was built). Here we use census block groups (CBGs) to approximate neighborhoods.

Other Data Sources

The validity analysis incorporated measures from three additional data sets describing the physical, social, and demographic characteristics of the neighborhoods of Boston.

Demographic characteristics were accessed from the American Community Survey (ACS; 2005–2009 estimates), including proportion of owner-occupied residences, median income, and racial composition.

Two measures of physical disorder, private neglect (e.g., unsatisfactory living conditions) and public denigration (e.g., graffiti), were derived from a database of requests for non-emergency government services, as compiled by Boston’s 311 system. Neighborhood-level measures were calculated in two steps. First, counts of all events in each category were tabulated. Second, they were adjusted using the neighborhood’s estimated response rate (see O’Brien et al., accepted). These measures were available for calendar years 2011 and 2012.

Collective efficacy was derived from the Boston Neighborhood Survey, a biannual telephone survey of adults in Boston neighborhoods with 3,428 participants in two waves (2008: N = 1,710; 2010: N = 1,718) recruited by random-digit dial. Sampling was conducted in a manner that guaranteed an approximately even representation across the city’s census tracts. Because tracts are larger than the census block groups that they contain, it was necessary to combine the two waves of the survey to provide sufficient participants in each census block group to enable measurement at that geographic scale. The BNS consisted of multi-item scales in which respondents reported on the social and physical ecology of their neighborhood, including a scale for collective efficacy, broken into two subscales capturing social cohesion and capacity for social control (Sampson et al. 1997). Scale items were averaged to calculate an individual-level measure. Neighborhood-level measures were then calculated as the average response of all residents, controlling for individual-level demographic factors.

Data Analyses

SAS 9.3 was used to construct parcel- and neighborhood-level measures from the permit database, and to conduct correlation and regression analyses using other neighborhood-level measures. To examine reliability at different time intervals, nested models were created with each CBG having a measure for multiple time windows (i.e., if 2 years of data were divided into 6-month intervals, each CBG would have four such measures). HLM 6.06 (Raudenbush et al. 2004) was used to compare the within-neighborhood consistency (intraclass correlation coefficient) and the reliability of measurement for five time interval settings: 1, 2, 3, 4, and 6 months.

Results

Content: Operationalizing Investment by Community Members

In order to create a neighborhood-level measure of investment, it was first necessary to develop a rubric for describing the project occurring at a particular parcel. Speaking broadly, construction could take one of four basic forms: demolition of an existing structure; new construction where there was no structure; addition to an existing structure; and renovation of an existing structure, with no change in its footprint. A given parcel was classified as undergoing one of these four as follows:

  • Demolition: A Demolition permit or a permit description of “demolition.”

  • New Construction: A permit for Erect New Construction, Excavations (as necessary for the building of a foundation), or a permit description of “new construction.”

  • Addition: A permit for Long Form/Alteration, Amendment to a Long Form, or appearance before the Board of Appeals, or a permit description of “addition,” all of which indicate extensive change to a structure. Because some of these might be included in demolition or new construction projects, a parcel was classified as undergoing an addition only if it was not already identified as undergoing a demolition or new construction.

  • Renovation: Any of the following permits indicating a modest interior modification: Short Form Building Permit, Gas Permit, Electrical Permit, Plumbing Permit, Use of Premises (i.e., minor changes to outdoor spaces), Cutting/Burning/Welding, or Asbestos Removal. Again, any of these might be part of a more extensive project, so a parcel was identified as undergoing renovation only if it did not qualify for any of the other categories.

In total, 45,092 parcels had requested one or more permits, of which 39,084 were attributed to one of these four categories; others had single permits related to people moving in or out or purchasing a building (e.g., Street Occupancy, Smoke/CO Inspection). The most common of the four categories was renovation, accounting for 70 % of projects (27,407). Second was addition, with 18 % (7,158). Not surprisingly, demolition (7 %; 2,738) and new constructions (5 %; 2,113) were much less common.

Renovations and additions would seem to be most in keeping with the theoretical construct of proactive investment, as they are most likely to be undertaken by property owners who are part of the community; the other two are in many cases being undertaken by outside developers (Logan and Molotch 1987/2007). To verify this distinction we calculated the proportion of parcels in each CBG undergoing each type of project and examined the correlations between these four measures (see Table 1 for all correlations and descriptive statistics). The strongest correlation was indeed between additions and renovations (r = .42, p < .001). The next strongest correlation was between demolitions and new constructions (r = .31, p < .001). Correlations across the two groups were weaker (r’s = .01–.28).

Table 1 Descriptive statistics for building categories and correlations between them
Full size table

The product of these analyses is an ecometric of investment by community members in their neighborhood from the records of building permits, calculated as the proportion of all parcels in a neighborhood undergoing a renovation or addition in the given time span. For the ~2.5 year period measured, the mean was 0.10 (SD = 0.05), meaning that in the average neighborhood 10 % of parcels underwent either a renovation or addition. Importantly, the measure can be distinguished from other types of building in a neighborhood. The proceeding analyses further probe the properties of this ecometric.

Reliability: Examining Consistency Across Time

With a methodology for measuring investment in hand, the next goal was to determine how long of a time period would be necessary to do so reliably. The database was divided up into a series of evenly spaced time intervals (e.g., 1 month), with each permit attributed to the time period during which it was issued. Any permit estimated to be granted after the last date of the database was excluded. Investment was then measured for each CBG for each time interval, generating a series of repeated measures nested within CBGs. For purpose of comparison, this process was done for 1-, 2-, 3-, 4-, and 6-month intervals.

We analyzed these nested datasets using multilevel models run in HLM (Raudenbush et al. 2004),Footnote 1 focusing on two main parameters. The intraclass correlation coefficient [ICC; neighborhood-level variation (τ0) divided by all variation (σ2 + τ0)] indicates the level of consistency in a neighborhood’s investment across time relative to other neighborhoods. Reliability (λ0) indicates how well this consistency can be used to discern inter-neighborhood differences. A linear parameter was included in the model to estimate the slope of change over time. It was allowed to vary across CBGs, giving rise to two analogous parameters that indicated whether the model could discern different trajectories across neighborhoods. Reliability was measured (λt), and instead of an ICC the significance of the level of variation in the slope (τt) was assessed using a χ2 test.

The reliabilities and ICCs from the multilevel models are reported in Table 2. All ICCs were significant at p < .001. Even for one-month intervals, 35 % of the variation was at the neighborhood level. By 6 months, this value crested 65 %, suggesting that at such an interval growth could be measured with reasonably high precision. It further suggests that an annual measure, which might be seen as best for most research, would be largely robust against measurement error.

Table 2 Intraclass correlations (ICC) and reliabilities (λ) for level (intercept) and neighborhood-level variations (τt) and reliabilities (λt) for cross-time change (slope) in investment across census block groups for various time windows, as derived from multilevel models
Full size table

Variation in slopes across CBGs was significant at p < .001 for all time windows except for 6 months (likely owing to having only 5 time points), indicating that the model was able to distinguish between the trajectories of different neighborhoods over time. The reliabilities state this would be most effectively done using one-month intervals. Though one probably would want a longer time interval for a one-time measure, these shorter intervals permit the model greater detail with which to estimate longitudinal trajectories.

Validity: Relationships with Other Neighborhood Characteristics

A final analysis sought to establish construct validity for the measure of investment via a series of correlations and regressions. We did this using the measure from 2011, as it is the earliest year for which concurrent measures of physical disorder were available, and therefore closest in time to the measures of demographics (2005–2009) and collective efficacy (2008–2010). These analyses were limited to CBGs that are primarily residential, excluding areas that might have commercial or institutional identities that have different dynamics (e.g., university campuses; final N = 428).

Initial correlations found that growth had the expected relationship with homeownership (r = .61, p < .001) and median income (r = .51, p < .001). Maps illustrating these correlations are presented in Fig. 1a, b. There was also a negative correlation with the proportion of the population identifying as Hispanic (r = −.22, p < .001), but no significant relationship with the proportion Black (r = −.04, p = ns).

Fig. 1
figure 1

Levels of investment across Boston’s CBGs, viewed alongside the distribution of (a) median income, (b) homeownership, and (c) overall disorder (a sum of private neglect and public denigration). Also highlighted are regions with greater investment than predicted accounting for its correlation with the other measure (a, b only). Legend is applicable to all three maps

Full size image

Investment had the hypothesized negative correlation with both types of physical disorder, public denigration (r = −.36, p < .001) and private neglect (r = −.34, p < .001; see Fig. 1c for a geographical representation). In order to determine whether these relationships were a consequence of shared demographic factors, or arose from some other behavioral or social dynamic, we ran a regression using demographics and physical disorder to predict growth (see Table 3 for complete results). Both public denigration (B = −.12, p < .01) and private neglect (B = −.12, p < .05) independently predicted lower levels of investment in a neighborhood. Unsurprisingly, the strongest predictors were homeownership (B = .40, p < .001) and median income (B = .31, p < .001). A higher proportion Black population positively predicted investment once controlling for other factors (B = .21, p < .001). The same was true for Hispanic population (B = .11, p < .05).

Table 3 Complete parameter estimates from multilevel models using demographics, physical disorder, and collective efficacy to predict levels of investment
Full size table

One possible explanation for this relationship between investment and disorder over and above demographic factors could be a neighborhood’s collective efficacy, which is often argued to influence behavior within the neighborhood, including maintenance of the public space (Sampson and Raudenbush 1999). Indeed, investment was greater in neighborhoods with higher collective efficacy (r = .35, p < .001). When collective efficacy was entered into the regression model (see Table 3), it maintained a significant relationship with growth (B = .12, p < .01). The relationship with public denigration remained about the same (B = −.11, p < .01), but the relationship with private neglect was diminished by about a third (B = −.08, p < .10).

Last, we wanted to do a preliminary analysis of how this measure might be reflective of gentrification. Gentrification is typically defined as the displacement of a poor or disenfranchised population by a more one with higher socioeconomic standing; thus, while our measure of investment is largely correlated with median income and homeownership, neighborhoods that deviate from this rule could potentially be undergoing gentrification. In Fig. 1a, b, we have highlighted CBGs whose level of investment was more than a standard deviation above the expected given the median income of and prevalence of homeownership in the community, respectively. The most prominent region in both maps is in two contiguous regions on the west side of the city known as Jamaica Plain and West Roxbury. Being that these areas are already in the highest quartile for socioeconomic status, this might reflect a continued popularity among young families looking to live inside the city.

Another cluster that is possibly more informative regarding gentrification is on the east side of the city in a region called Dorchester. A series of CBGs running in a diagonal line from northeast to southwest had greater than expected levels of investment. They run along a major thoroughfare (Dorchester Ave.) known to be a significant demographic boundary, with working-class white communities living to the east and poorer black and immigrant communities living to the west. Nearly all of the neighborhoods with elevated investment were to the west of this line. It is noteworthy that these neighborhoods were not in the lowest quartile for socioeconomic status, possibly reflecting the known pattern that gentrification occurs in neighborhoods of lower socioeconomic status, by definition, but not in the most disadvantaged neighborhoods. These interpretations should be treated with care, however, as the analytic approach is preliminary.

Discussion

The analyses here constituted the development and evaluation of a methodology for translating a database of building permits into an ecometric of proactive investment by community members across urban neighborhoods, including criteria for reliability and an initial examination of construct validity. Within this bigger picture, three things in particular stand out.

  1. 1.

    The richness of the data made it possible to construct a measure that was more than a mere tally of permits issued. We leveraged the geographic precision to make discrete projects at specific properties the basis for measurement, and used the detailed classification and description of the permits to exclude large developments, likely driven by extra-community forces (Logan and Molotch 1987/2007). In sum, this led to a measure that was more robust in its focus on local processes.

  2. 2.

    The reliability analysis indicated that investment could be measured every 6 months for CBGs. Previous work combining CBG and tract analyses would suggest that the same might be true for two- or three-month periods for tracts (O’Brien et al., accepted). As a point of comparison, a single whole-city survey or observational audit typically requires longer than that to complete, meaning repeated measures of this sort are unprecedented in ecometric research. This enables longitudinal analysis of cross-neighborhood trajectories in investment.

  3. 3.

    The final stage of the analysis not only demonstrated basic validity, in that investment was most associated with homeownership and income, but also highlighted it as a counterpoint to disorder, constituting the other end of a spectrum of neighborhood maintenance. Importantly, though this relationship can be partially attributed to the role of homeownership, median income, and collective efficacy in driving both disorder and investment, but still persists independent of these factors.

To summarize, the work here provides a novel ecometric that could shed new light on neighborhood dynamics and processes. While theoretically related to one of the classical measures in neighborhood research, physical disorder, it is particularly distinctive because rather than yet another indicator of blight, it offers an indicator of community thriving, something rarely seen in the field. Further, as with other measures derived from large-scale administrative data sets, it is inherently longitudinal, opening the door for a number of advanced methodological techniques. These might include cross-lag models that examine relationships between various neighborhood characteristics across time (Kenny 2005), and predictive analytics that use localized patterns to predict the likelihood of future events. With all of this in mind we explore a few ways that this measure might be utilized to advance research on urban communities.

The most intriguing question is the social or behavioral dynamics that link investment and disorder. We propose three possible explanations. The first and simplest is that this linkage can be attributed to an additional unobserved variable, apart from those demographic and social measures included in the analysis here. Though possible, this seems the least probable as the analyses here included many of the measures that would be considered most likely to play this role. Second, it might be that investment and physical disorder are part of the same behavioral construct; individuals who are inclined to invest in the neighborhood are also less likely to create disorder, and possibly more likely to take action to prevent or eliminate disorder. Thus, neighborhoods whose residents are on average higher on this construct will see both greater investment and lower disorder. A third interpretation is that one or both measures have a causal influence on the other. For example, disorder might signal a failing neighborhood, discouraging other residents from investing, a model that would fit closely with current perspectives on the psychological impacts of disorder (O’Brien and Wilson 2011; Ross et al. 2001). Alternatively, investment, by demonstrating care for the neighborhood on the part of some, might inspire others to take better care of the neighborhood, or make them less inclined to denigrate it in some way (Keizer et al. 2013). A reciprocal relationship might also be possible.

A second potential area for study regards the influence that patterns of neighborhood maintenance and deterioration can have on behavior, particularly when these responses perpetuate or counteract the expansion or escalation of disorder and other, more serious problems. Existing work has repeatedly demonstrated how people use physical disorder as a heuristic for evaluating neighborhood quality and safety, leading to important behavioral consequences (O’Brien et al. 2014; O’Brien and Wilson 2011; Perkins et al. 1992). There is also a growing literature on how these same perceptions and conditions can have biophysical impacts, leading to elevated stress, depression, and lower overall mental and physical health (e.g., Furr-Holden et al. 2012; Kruger et al. 2011; Theall et al. 2013; Wen et al. 2006). It is an open empirical question, in that case, whether the presence of investment, as a signal of thriving in a community, plays a countervailing role in these same areas. Does it generate visible artifacts that residents and visitors alike can observe and interpret as symptoms of a healthy community? Does it have ameliorative effects on perceptions and psychological health where disorder has negative impacts? Depending on the results of these and allied lines of inquiry, it would be possible to shed light on the bigger question: where disorder is central to cycles of decline, is investment just as important in a neighborhood’s upward trajectory?

A third opportunity for future research is how the measure of investment might be combined with other information in order to capture patterns of gentrification. Gentrification has been a major area of interest, with researchers attempting to understand the patterns by which it arises (Hwang and Sampson 2014) and the impacts that it can have on the individuals and communities that are already present (Formoso et al. 2010). Nonetheless, there is no standardized index or set of indices for measuring it. Gentrification certainly takes multiple forms, but it might be argued that new building projects are nearly universal across them. We have used a mixture of statistical and visual analysis as a preliminary approach to using this measure to identify regions of ongoing gentrification. Interestingly, the approach we took—highlighting neighborhoods with levels of investment greater than would be expected considering the local levels of median income and homeownership—exposed regions of both high and low socioeconomic status. It is an interesting question as to how the patterns of change in these two types of neighborhoods are similar and how they differ, and which would be properly referred to as “gentrification.” That said, building projects are certainly not the only component of gentrification, and the measure here would merely be a part of a more comprehensive effort to quantify and track patterns of neighborhood, one that will necessarily involve direct measures of demographic change and potentially other indicators of shifts in the physical and social ecology.

Last, a measure of investment would be useful for policymakers, practitioners, and other community members and leaders looking to identify and address the needs of urban communities. Though these data are already public record and therefore accessible (depending on a particular city’s open data policies), the information they provide in their raw form can be difficult to interpret. The methodology here has given structure and meaning to the data and provided an accessible metric that describes a specific dimension of a neighborhood’s ecology. While our focus throughout has been primarily on long-term patterns and broad, cross-neighborhood comparisons, those who serve and live in neighborhoods can use the same information in a more targeted fashion. Community groups and practitioners could use these and other measures to track dynamics of blight and thriving in a neighborhood. They can also dig down and look at address-level conditions, noting specific addresses or blocks that are most contributing to shifts in the overall appearance of the neighborhood. This knowledge could be translated into more effective communication with the community, and could inform current services and interventions or inspire new ones. In order to support such efforts within Boston, the Boston Area Research Initiative, the research program through which the current study was conducted, has made these measures publicly available through an online mapping platform called BostonMap (worldmap.harvard.edu/boston/). Importantly, such efforts need not be separate from research. If interventions are implemented as experiments, the continuous nature of the data greatly facilitates their evaluation by automatically providing both before and after measurements. In these ways such data could not only support both social science research and community practice, but also collaborations at their intersection.

In closing, it is important to consider the limitations and challenges that face work based on this measure. Most importantly, researchers will need to acknowledge and address the assumptions that come with using an administrative data set. Because the data are not collected as part of a research protocol, their meaning and interpretation must be inferred, requiring careful attention to the information that the data contain, but also to the process by which they are generated. This process is of critical importance; not only does it give the data set its basic meaning, it can also be the source of systematic bias. For the case here, as noted at the beginning, residents do not necessarily apply for permits for all work that they might do on their property. Such evasions would be most common for small projects, thus the measure here is likely less precise in capturing these low-end investments, and therefore has greater measurement error for neighborhoods where such projects are more common. On the positive end, it is not clear how this would be responsible for any of the main findings here, rendering them artifactual. In this sense the measurement error for this measure is unlikely to be greater or more concerning than more traditional methodologies, like surveys or observations, which have their own difficulties to overcome (i.e., reporter bias, inter-rater reliability). Nonetheless, future work should always be conscious of the assumption that this measure faithfully reflects all investment in a neighborhood.

For each of these areas of research, the opportunity is not only for single-city analysis, as we have done here for Boston, MA, but for cross-city comparisons. It is feasible that the same measure could be created for any city that documents building permits. This task, however, might not be as straightforward as it appears. Cities differ in the types of work for which they require permits, and the names by which they refer to them. They might also have locally specific processes for documenting permits, giving rise to databases whose information does not have the same exact interpretation. For research in other cities using permit data to proceed, the concepts here would have to be translated to the parameters of that city’s database. More importantly, cross-city comparisons would require methodologies that assess the comparability of these measures in different cities. Such work will be necessary for the full opportunity that large-scale administrative data present for community research to be realized.

Conclusion

The methodology here provides a new way to measure and study proactive investment by community members in the long-term upkeep of a neighborhood, making it a counterpoint to the canonical measure of disorder, which reflects failures of maintenance. It is distinctive in its emphasis on community thriving rather than blight, giving additional potential to advance research and practice regarding both upward and downward neighborhood trajectories. From a broader perspective, it is also an illustration of the ecometric opportunity offered by “big data” in the form of large-scale, digital administrative records. Building permits are just one of many such data sets, the sum of which could generate dozens of different measures of neighborhood conditions and patterns, spanning concepts both traditional and novel, capturing both blight and thriving. Though there is a need for care and robust methodology when working with such data, the potential contribution appears to merit further exploration and study.