Understanding place and health: A heuristic for using administrative data

Abstract

The increasing availability, use and limitations of administrative data for place-based population health research, and a lack of theory development, created the context for the current paper. We developed a heuristic to interrogate administrative data sets and to help us develop explanatory pathways for linking place and health. Guided by a worked example, we argue that some items in administrative data sets lend themselves to multiple theories, creating problems of inference owing to the implications of using inductive versus deductive reasoning during the research process, and that certain types of theories are privileged when used administrative data bases.

Introduction

Recently public health has witnessed a proliferation of studies investigating the relationship between various attributes of places and the health of their populations (Kawachi and Berkman, 2003; Pickett and Pearl, 2001; Tunstall et al., 2004). In search for answers to the difficult question “How does place shape and influence health status at a population level?” researchers concerned with this issue are exploiting data from an increasingly vast array of administrative sources such as records provided by police, schools, social welfare agencies, park administrators, and day-care facilities, to name just a few (Sooman and Macintyre, 1995; Ellaway and Macintyre, 1996; Yen and Kaplan, 1999; Macintyre, 2000; Giles-Corti and Donovan, 2002; Cummins et al., 2005).

Using administrative data is a convenient way of undertaking research. First, population health data such as vital statistics and health services utilization data are widely collected and increasingly easy to access. Second, these data are particularly appealing for studies of health and place due to the ability to link them (albeit with difficulty) at multiple scales to other population sources of information such as socio-demographic data originating from censuses (usually using some common or compatible geo-coding, like postal codes and census tracts, which may allow for re-aggregation to different scales). Given advances in computer capacity, many researchers now have a powerful arsenal for empirical research that has significantly enabled linkages between various data sources.

These data, however, have not turned out to be quite the panacea one might have expected. Studies of health and place are now increasingly fraught with the issue of what to do with this data. How exactly can we explain its relationship to health outcomes at an aggregate level, for instance? What data is missing, and why? How can we draw specific inferences from generic measures in secondary data? What problems arise when using secondary data for a purpose other than the one for which it was collected? This paper addresses precisely these issues. The use of administrative data bases in health and place research is inherently an inductive endeavour. Data is accessed, analyses are run, and explanations are sought. We propose a devise that permits for a less purely inductive process to drive future research.

The largely inductive process encouraged by an over-reliance on analysis of administrative data has been discussed as problematic in the current social epidemiological literature (Pearce and Davey-Smith, 2003; Muntaner and Lynch, 1999; Krieger, 2004). For instance, both Pearce and Davey-Smith (2003) and Muntaner and Lynch (1999) point out that most ecological studies of income distribution and health have not been largely guided by theoretical propositions, but rather, have relied nearly exclusively on secondary data analysis alone. While these studies have brought about some statistical evidence that income inequality is positively associated with national mortality rates, the under-theorization of these studies have led to an enormous amount of post-hoc speculation as to exactly which mechanisms are responsible for the empirical results. As a result there has been little agreement as to the explanations for these patterns (that is, how they come about) or what they may mean for social policy. A purely inductive use of data, largely unguided by theory, is partly responsible for this confusion.

Given that data from secondary administrative sources normally provides only single indicators (such as “race” or “SES”), research that borrows from these sources has to rely on these single indicator relationships, and thus, their use is limited. This hampers research if the goal is to go further than merely showing empirical associations and to explore the social mechanisms mediating between social phenomena and health outcomes. Krieger et al. (2005) note “race” in the United States as a flagrant example of this problem. They state that most public surveillance systems in the United States, if they have collected data on “race” at all, have not then collected data on other related socioeconomic factors, thus limiting our potential understanding of the relationship between economic and non-economic aspects of racial discrimination.

It is therefore worth recognizing that administrative data sets were designed with an original purpose (e.g.: client tracking, forecasting, payment, accountability, legal requirements)¹ that is often quite different from our own research interests. This does not mean that administrative data are inadequate for research on health and place, but that we should be more critical about the extent to which our use of this data may privilege some types of explanations over others.

The increasing availability, use, and limitations of administrative data for place-based population health research created the context for the current paper. The problem of sorting out how to link existing data to understandings of the association between health and place emerged during a seminar between two research groups in two Canadian cities. Problems began to arise when we realised we had enormous amounts of rich and diverse “data”, but no theoretical framework within which to place it. How should we link, both conceptually and methodologically, such a huge and varied inventory of information? How might, for instance, proportion of library books borrowed per month, or proportion of speeding fines per head of population stack up against each other as measures of social cohesion? How might each explain health outcomes? Furthermore, if we intended to provide recommendations for the collection of new data, how would we identify the gaps? What new data might be needed and why? Discussions within the public health literature appeared to be somewhat scant on this issue, partly because our understanding of how to move from theory to measurement is generally underdeveloped in this field.

Essentially these problems boil down to an extensive use of administrative data in the study of place and health, accompanied by an under-utilisation of theory in relation to health–place linkages. Theoretical considerations are critical as they describe the putative causal linkages between health and place, which in turn, tells us where, and perhaps how, we might best intervene. Given that few empirical research studies with a focus on health and place are explicit in their assumptions about causal mechanisms, or in their theoretical process, we have a limited capacity to explain how place affects health. This lack of clarity may lead to interventions and policies that are ineffective or inappropriate if they are based on etiologic research of this sort.

In striving towards a more self conscious and reflexive use of administrative data sources we turned to some of the basic lessons afforded to us by measurement theory in psychology and education, as well as the basic principles of variable operationalization used in the social sciences (Crocker and Algina, 1986). Applying these basic notions to the problem at hand, we adopted a heuristic to interrogate administrative data sets and to help us locate particular information (such as new secondary source materials regarding information about vandalism or traffic congestion in neighbourhoods for example), along an “explanation pathway” for linking place and health. Part of the role of the heuristic was to enable us to appraise the appropriateness of the items from administrative sources with regards to particular place-specific theories about health.

The aims of this paper, therefore, are: (1) to develop a heuristic that could help us understand how theory and administrative data can be better used when attempting to explain the links between health and place and (2) to illustrate the use of this heuristic device. We call this heuristic device a “template”, a guide and tool for researchers to help them critically examine their own research. Using the template as the guide, we seek to meet three objectives: (1) to demonstrate that some items yielding from administrative data sets lend themselves to more than one theoretical explanation; (2) to examine some of the implications of using inductive versus deductive reasoning during the research process; and (3) to illustrate that certain types of theories are privileged by our use of administrative data bases.