510 forum

Problems in Identifying Public and Private Organizations: A Demonstration Using a Simple Naive Bayesian Classification

Jonathan Rauh

Published online: 6 November 2013 # Springer Science+Business Media New York 2013

Abstract Publicness theory has received considerable treatment over the past 20 years. While the theory has provided much in the way of theorizing on how to think about public and private organizations, it has also raised unanswered questions. A major question in this regard is, given this theory, how should we go about classifying organizations as public or private given both what is known from the theory as well as an organizations legal definition. This manuscript seeks to address this question by identifying the components necessary for a classification scheme.

Keywords Publicness . Classification . Organizational theory

Introduction

How do we categorize organizations as public or private? A great deal of thought has gone into this notion starting arguably with Dahl and Lindblom (1953), but most certainly with Bozeman (1987). The number of organizations providing services used by the public is large and increasing (Anderson 2012). Despite this there is not much beyond theory in classifying organizations as public, private, or some combination thereof. By this I mean it is understood that classifications using political versus economic authority are important1, but they do not tell the whole story—a point already made by others such as Perry and Rainey (1988); Bozeman (1987); Andrews et al. (2011), Meier and O’Toole (2011) and others. Identifying means of categorization between public and private is a perennial question whose importance is rooted in the fact that we as a society often impose public purposes on private organizations (Perry and Rainey 1988); that management techniques are transferred between sectors (Meier and O’Toole 2011); and that motivations for choosing to work in one sector or the other

Public Organiz Rev (2015) 15:33–47 DOI 10.1007/s11115-013-0250-y

1This was operationalized by Emmert and Crow (1988) as the percentage of the budget derived from the public sector.

J. Rauh (*) Department of Political Science, University of South Carolina, Columbia, SC 29208, USA e-mail: [email protected]

are still not clearly defined (Perry et al. 2008; Perry 2000). This is not meant as a criticism of the important work that has been done and is currently being done on publicness – e.g. Bozeman and Bretschneider (1994), Miller and Moulton (2013), Bozeman (2013) and so on. Rather it is meant to examine the way that we think about the publicness or privateness of organizations based upon the existing criteria for classifying organizations as more public or more private – a call that has recently received increased attention from Bozeman (2013) in his article detailing the need to address the link between public policy and publicness.

Bozeman and Bretschneider (1994) state that organizational goals are indicative of publicness. However these organizational goals are just proxies that point to something greater—public versus private values. Bozeman (2007) clarified this distinction as public sector values being inputs that are used for public sector outcomes. This offers a higher degree of leverage to measure publicness but it still does not quite provide a means of distinguishing what is public and what is private. This is because even outcomes with profit motives can serve a public purpose – e.g. the recent work by Miller and Moulton (2013) detailing the degree to which private firms that serve public purposes enter or leave the market based upon the proportion of public providers in the area. In this paper I argue that if one were forced to classify an organization as public or private based upon the set of available criteria for publicness, they would have a difficult time making hard distinctions. In doing so I claim that the current use of individual criteria is not satisfactory for classifying organizations as public or private since individual criteria only capture traits of organizations. What is needed instead is classification scheme that treats the set of available criteria as interworking. This is not necessarily a new argument, but most classification schemes used to identify public versus private have generally been taxonomies or ontologies.

The reason for more discrete classifications is simple—distinguishing between public and private is complicated. As evidence of this there are few studies which operationalize publicness along multiple dimensions, and even these studies focus on single organizational types—see for example Bozeman and Bretschneider (1994) and Miller and Moulton (2013). In particular the Miller and Moulton piece is informative because it details the effects of external, non-stochastic, variations in the provision of public services, i.e. in areas with more public service provision, private organizations are less likely to provide goods and services with a public mission, especially when there is a clear economic loss. These findings are counter to the proposals put forth in more strict political economy arguments such as those from Charles Wolf (1988). This is not a criticism so much as recognition that like theories from political science, publicness has good formal attributes that are more difficult to operationalize. When these theories are operationalized with greater specificity, it is generally found that the specifics contribute as much if not more than the core idea they are meant to convey, e.g. Bozeman and Bretschneider (1994). Using the current framework for publicness I imagine several organizations along the various spectrums of publicness. Using a naïve Bayesian classification I demonstrate that while it is true that publicness involves more than just ownership, other dimensions of publicness are necessarily non-random as a result of this prior necessity.

Much has been written on the lack of rigor in public administration and political science in general as compared to the “hard sciences” (Miller and Yang 2007). Various suggestions have come out of conferences and publications on how this should be addressed. For example, Lewis and Grimes (1999) claim that researchers should examine alternative

34 J. Rauh

paradigms in order to understand contrasting explanations of the phenomena at hand. But this still do not solve the problem of multiple interpretations. For example, Warner and Hefetz (2010) examine reasons for decreases in private contracting in local governments post-1997. They find that managerial learning and an appreciation of citizen expectations necessitated increased involvement of public management and hence more public-private relations as opposed to simply private. But what of an alternative view proposed by DeHart-Davis and Kingsley (2005) that public managers can be captured by contractors and who can then transfer the majority of risk to the public sector? These represent distinct although not entirely opposite views of a similar question. So how does one judge which is more correct? Do we simply examine who has the better methodological framework? Perhaps, but this does not necessarily get at which theory, or which model, is more correct. As noted by Clarke and Primo (2012) theoretical models are never fully parsed out with actual data (the real world is just too messy). They are tested with hypothetical data that are notoriously fragile, e.g. the conflict is not between theory and data but between different models that never fully capture the theory and all of which are flawed. Clarke and Primo (2012) go on to argue that models are akin to maps. No one would really suggest that a map is the world; it is just a 2D representation of a 3D world.

Other scholars such as Flinders and John (2013) are more nuanced in their criticism while also noting the social impacts of depoliticizing political science in favor of methodology. They do not argue against advanced methods, but rather argue that political science must remain relevant to politics and society. When advanced methods are used, there must be keen attention to translating the findings in a manner that is readily accessible. In this paper I apply the same argument to publicness theory. Although I would argue the field in general could do with more methodological rigor, I do not mean to say that we should turn to methodology above all, or that theory building should not be the most important part of the work. In addressing how we define organizations as public versus private I recognize that there are not easy categorizations and at times the grey area can limit theoretical insight. But this does not mean that we should simply turn to methodological rigor in order to address the grey area, nor does it mean that we should simply use cases as one-offs in pursuit of the theory. To demonstrate this I define the problem of public private classification as detailed from Bozeman on. I then use a naïve Bayesian classification based upon the various components that go into publicness. The question at hand is, “knowing that publicness comprises these factors, how would we define an organization along the public-private spectrum based upon its representation within these categories?”

This paper proceeds as follows: a discussion of the current thinking on classifying public and private organizations; a discussion of how a dataset for measuring publicness was created and a discussion of naïve Bayesian classification; and a finally results, conclusions and discussion.

Ways of Classifying Public and Private

Despite the significant study of public-private differences there are still not definitive answers to the question, “are public organizations different than private organizations?” The original concept of publicness was described on a continuum via a two-dimensional frame (Bozeman 1987). However the application of the theory has required significant

Problems in Identifying Public and Private Organizations 35

revision. Meier and O’Toole (2011) claim that to address publicness in the current parlance one must think of it as a menu of characteristics. I agree with this point on two grounds. First, as pointed out by Meyer and Williams (1977), Andrews et al. (2011) and Bozeman (1987), comparing public and private organizations is an extremely complex problem. Second, and party to the first point, specifications that work well for one organization may not work well for another organization despite similar purposes and designs. The menu of characteristics including a public purpose and the instruments and strategies used for achieving that purpose has traditionally defined organizations as public or private. If we consider the multiple models presented by Andrews et al. (2011) then the distinctions become readily apparent and are made more-so by a statement from Andrews et al. (2011: i285) that sums up the problem nicely,

The distinction between a public purpose and the instruments used to implement that purpose (ownership, funding, control) is important because what is defined as a public purpose can change, but the policy instruments can persist and thus reflect dated decisions about publicness.

A ready example of this is General Motors Company. On July 10, 2009 General Motors emerged from Chapter 11 bankruptcy as General Motors Company with an initial public offering and the US Government taking a 61 % stake in the company. A private board of directors still ran the company though and federal regulations prevented shareholders from touching assets in Europe and Asia. So what was General Motors Company—public or private?

From a discussion about the organization’s public purpose the answer would be, “maybe;” and from the discussion of the instruments used the answer would again be, “maybe.” The purpose of the company is profit driven but the purpose of the stakes take in the company by the US was to prevent industrial decline on a scale that clearly served the public’s interest. So depending upon the tact one takes as to company’s purpose they could claim that the profit motive shored up a public interest in keeping GMC afloat.

As Rainey (2009) notes, a public purpose may not differ across sectors, but the mechanisms (ownership, control, and funding) do. The facets of ownership, financing and control become more complicated in this story. After all, it is possible to separate ownership from control. As Stout (2002 and 2011) notes, shareholders do not really control a company, despite the popular myth. We have multiple examples in which a board of directors goes against the wishes of shareholders—sometimes with good consequences, sometimes with disastrous consequences. On the public side, taxpayers are typically not organizational theorists or public managers. They do not get to walk into an agency and tell the manager how to best deliver public services—note that there are some instances of collaborative management and evaluation but this still deals in delegated authority and managers are not legally or otherwise formally bound to act upon the wishes of the general public.

Different Ways of Considering Public-Private Distinctions

The traditional property rights argument assumes that the practices of a public organization are different from a private organization because the public organization has no profit motive (Alchian 1977; Clark 1973), e.g. there is not a legitimate external threat of pain for a public organization to alter its behavior. I take this argument to be

36 J. Rauh

too simple. Public organizations face similar threats of pain, just not as directly from their customers (wards)—see Tompkins and Jos (2009). If a public organization underperforms then its budget may be hit or in some cases its duties may be taken on by another agency. But this still misses the bigger picture. It is not just that the values of public and private organizations are distinct, i.e. that they care about different things. There are differing scopes of control in public and private organizations as well as other empirically verifiable realities.

The public sector serves groups and individuals with more restricted choices. The result is of course that the public sector takes on multiple roles, not just that of provider or seller (Tompkins and Jos 2009). It takes on the role of auditor, compliance manager, quality manager, etc. As such judgments about the quality of provision cannot be so neatly pinned down in distinctions between public and private. Despite this, the public sector has borrowed a great deal in the way of performance measurement and performance management from the private sector—largely a result of calls to run government like a business (Moynihan and Pandey 2010).

This is interesting given the reference to quality. While the public sector may be the arbiter of at least a minimum level of quality, it may measure the attainment to, or beyond, that level using techniques that were developed in the private sector (Moynihan and Pandey 2010; Moynihan 2009). This gets back to the efficiency argument that both the public and private sector are beholden to. The public sector has a mission and an incentive to provide a restricted series of choices to the public whereas the private sector is in competition to provide choices that are either better quality or less expensive (or both) than their competition. As such they have an incentive to have more accurate information. The public sector, because of its restricted provision does not have the same incentive to innovate in performance measurement, but to use the performance measurement techniques that have been tested and proven (Moynihan and Pandey 2010; Moynihan 2009).

There is a flip side to the previous argument made about private sector entities using publicly tested techniques to provide public services. If it is not an engineered solution that they have in hand (such as better garbage trucks, better tracking mechanisms, etc.) then the private sector tends to use techniques employed in the public sector for public service delivery. Of course this is only true for engineered problems, not for social problems. Jones and Baumgartner (2005, 2012) point out this distinction between engineered problems, those with a set and logistically verifiable solution (such as mail delivery) versus social or sticky problems, those with multiple facets to a solution (such as solving poverty or education). Although not tested here explicitly, it seems from much of the literature that when addressing engineered problems the private sector may indeed outperform the public sector, or the public sector may look to solutions for service delivery from the private sector. However for social problems (such as how to educate a nation or solve poverty) if the private sector wants to get in the business of social provision, then they will tend to rely on solutions offered by the public sector. For example, the pedagogical techniques used by private schools tend to be the same as those used by public schools (Gill et al. 2001; Gaskell and Levin 2012). Additionally, healthcare tends to be delivered in the same manner in public hospitals as in private hospitals (Anderson 2012).2 So, just as private sector firms may emulate their more

2 A possible exception is in the delivery of specialty treatments that tend to be more prevalent in private hospitals.

Problems in Identifying Public and Private Organizations 37

successful rivals, the public sector has a habit of emulating successes in the public sector3.

The above is important because public organizations address problems for which the efficiency frontiers are not specified just in monetary utility (Williamson 1999); additionally concerns over monetary utility exclusively can lead to inefficient transactions and the types of solutions proposed by economists are not necessarily politically expedient because they may be too transparent (imposing costs on a geographically specific population to pay for a project with a general benefit is the classic example used by Arnold 1990). Furthermore, the types of activities engaged in by government can often increase the efficiency for the whole and avoid externalities (think regulation or the production of highly specified materials that the market will not produce—The Manhattan Project for example).

A final point is that of managerial motivations. The goals of organizations and of managers in private and public organizations can be both reflexive and transitive. Consider the case of Chic-Fil-A. They are a for-profit, privately owned organization. However, they make a choice not to open on Sunday. They have the capacity to make greater profits if they were open on Sunday, but they have chosen to project a certain value. Likewise, the actions managers take, and therefore the actions organizations take, are reflective of the organizations values. Managers in the private sector want the “corner office” but managers in the public sector are also motivated by money and prestige as well. Even in their studies of public service motivation Perry and Wise (1990) and Finlay et al. (1995) do not make the case that public service motivation is anathema to desire for reward and recognition.

Implicit in this discussion of publicness in performance measurement is the question, do public and private organizations care about different things (effectiveness, efficiency, equity, etc.)? The answer implied form the examples given above would seem to be, “sort of.” There are a variety of studies of publicness consisting of varied methodologies which come to similar conclusions: public and private both have a value for effectiveness and efficiency, but public organizations also have a value placed on equity and service delivery (this has been noted in drug abuse prevention, abstinence, etc. – Heinrich and Fournier 2004; mortgage lending – Moulton 2009; and other areas). Again though this argument cannot be completed as a public-private distinction, especially given the example of Chic-Fil-A above. So differences exist, but so do similarities—where the decision point lies may then be a sectorial factor.

Categorizing Organizations as Public or Private from a Methodological Perspective

The problem with classifications is that they tend to say little about the cascading effects of being classified as one thing or another. In other words changes to the left-hand side of a rule are attributed to changes in values listed on the right hand side. Once I classify something as public, but it starts to take on the attributes of a private organization, how

3 This has an effect for organization structure and culture which will be discussed below—findings from Frumkin and Galaskiewicz (2004) suggest that where an organization chooses to focus (internally or externally) has effects for its organizational make up, regardless of whether it is public or private.

38 J. Rauh

does that affect my classification? What, for example, would one use to classify GMC in 2013 as opposed to 2010? Profit motive? Organizational performance? Ownership?

Reiterating an above statement, “…comparison of public and private organizations is fraught with complexity” (Meyer and Williams, 1977). What is also apparent though is that we are dealing with characteristics that are largely separable. We can separate ownership and control, e.g. Stout’s (2002 and 2011) argument considering the myth of shareholder dominance. After all a shareholder, even a majority shareholder does not walk onto the floor of a factory and tell them how to produce a truck any more than a taxpayer walks into the public research institute and tells a scientist how to approach a research topic.

What is left is a series of criteria that are separable, but all of which work together in some form or fashion to create what we classify as a more public or a more private organization. There are expectations as to how these criteria “should” interact based upon previous experience or popular expectation, but nothing entirely clear cut. As Andrews et al. (2011: i317) state, “The consequence of these substantive and methodological problems is that it is impossible to conclude with any confidence that publicness makes a positive or negative difference to organizational performance…” My position is that while hard and fast classifications may be difficult to come by, we can look at organizations based on the characteristics that it exhibits and from there determine if those characteristics match up with the organization’s legal definition as a public or private organization. To do this I suggest the use of naïve Bayesian Classification.

The advantage of this technique is that instead of aggregating publicness to a spectrum measurement, or placing purposes and risk/reward into quadrants, Bayesian classification addresses the dimensionality of publicness by treating each dimension as independent and then examining the probability of seeing a particular dimension’s value (or range of values) given a prior classification. In other words, once we figure out the probability of a characteristic belonging to a particular classification, we simply pick the most probable classification.

Naïve Bayesian Classification

Naïve Bayesian Classification is straightforward and it provides a flexible means of dealing with multiple attributes. It has been shown to deal remarkably well with a wide variety of problems in spite of its simplicity. I describing how this method may be applied, similar to Train (2003), I assume that the attributes used to describe any outcomes are partitioned into stable and flexible. This suits with the idea that the attributes of public and private organizations are independent of each other4. Values with flexible attributes are changeable whereas values with fixed attributes are not—so fixed versus variable effects. So the date an organization was founded is stable. The degree to which an organization uses strategy a or strategy b is flexible. To identify categories for comparison one must first identify the potential outcomes of being public, private or mixed. This is defined as a term [(y)∧(a→b)]⇒(c→d), where y is a conjunction of fixed conditions shared by both outcomes since no organization is 100 % public or 100 % private, (a→b) represents the

4 Note that if a feature is correlated with another feature then it will overweight the contribution of that feature. This can be easily addressed by weighting the feature by the covariance between the correlated features which ostensibly removes obsolete or redundant information from the dataset—see Ratanamahatana and Gunopulos (2002).

Problems in Identifying Public and Private Organizations 39

proposed changes in values of flexible features, and (c→d) is the desired effect of the action. So for example a government entity can pursue private bonds with the intent of increasing it performance on some predefined measure which benefits the public. Thus altering the degree to which is publicly financed, with the intent of achieving some goal that is good for the public sector. This provides an estimation of how much some attribute(s) needs to be changed in order to effect some change of moving undesirable outcomes towards desirable outcomes.

Bayesian classification relies upon Bayes theorem, which states:

p xi � � �y

� � ¼

p y � � �xi

� � p xið Þ

p yð Þ where

& p(xi|y) is the probability of instance y being in class xi, e.g. what we want to compute; in this case it is the probability that an organization is classified as public or private and has some characteristic xi.

& p( y|xi) is the probability of seeing y given characteristic xi. From the literature and from experience we know that both public and private organizations share features, so what is the probability that we see those features in a given organization?

& p(xi) is the probability of occurrence of feature xi, i.e. just how frequently does xi show up?

& p( y) is the probability of occurrence of being y; so how often is it a private organization versus a public organization (legally defined)?

An example: Assume we have two classes of activities, xi&xj where xi = an organizational belief

that improving performance measurement techniques can increase organizational efficiency, and xj = an organizational belief that performance measurement techniques are merely tools for evaluating performance. We have an organization y whose characteristic as public or private is unknown. Classifying y as public or private in this instance is the same as asking, is it more probable that y is public or private.

To estimate whether an organization is public or private I used a naïve Bayes characterization. Naïve Bayesian Classification is very useful in separating out items with high dimensionality. Given the multiple criteria required to address the public- private distinction, this may as a better means of classifying organizations as more public or more private than just taxonomies.

If we have some data on the types of values espoused by organizations and their legal classifications, then we can see how this works—see Table 1.

p xi � � �Private

� � ¼

2

4 � 4 7

2

7

¼ 0:5 � 0:57 0:29

¼ 0:98

p xi � � �Public

� � ¼

0 � 4 7

2

7

¼ 0

40 J. Rauh

So in the example above, it is more likely that when we see an organizational belief that improving performance measurement techniques can increase organizational efficiency, that organization will be private. This can be simplified by assuming that features have independent distributions, e.g.

p y � � �xi

� � ¼ p y

� � �x1

� � � p y

� � �x2

� � � … � p y

� � �xn

� �

The naïve Bayes approach simply assumes that certain features exist which classify something as being in one group or the other (Rish 2001; Hand and Yu 2001). Given that much of what defines public versus private deals with description this seems to be appropriate. From the literature it is known that these descriptive features fall into the categories of Managerial Motivation (M), Intended Outcomes (I), Performance Measurement (Q), Values (V), Ownership (O), Control (K) and Financing (F).

The naïve Bayes characterization would then be p(CPu,Pr,PP|M,I,Q,V,O,K,F ) Using Bayes theorem this implies:

p CPu;Pr;PP � � �M; I; Q; V; O; C; F

� � ¼

p CPr;Pu;PP � �

p M; I; Q; V; O; K; F � � �C

� �

p M; I; Q; V; O; K; Fð Þ To measure C for initial classification use the legal arrangement under which it

operates. Is it a privately held of publicly held (stockholding) company? If so then classify it as CPr Private. Is it a Public Agency in a legal sense? So is it an executive agency, a legislative bureaucracy, a public service entity, etc.? If so then classify it as CPu Public. Is it a public-private partnership? If so then classify it as CPP. For the variables on the right hand side, assume that these can be scaled on a spectrum from Private (0) to Public (1). It may seem that the means of scaling these items is dependent upon the paradigm through which one views public administration. Does one take Morgan’s (1980) functionalism, radical structuralism, etc. language or do they take one of the multiple paradigms of Babbie (2005)? So do we measure these items based on how they operationalize within each setting or do we measure them based on a theory about what is public from some already established theory? Although public administration as a field lacks clear classifications because of its multidisciplinary nature (Miller and Yang 2007), in attempting to categorize publicness one is already dealing with multiple classifications of financing, managerial motivations, etc. As such attempting to distinguish public from private (or demonstrate a lack of distinction) is a question of both measurement and practice. Still the distinction is meant to distinguish

Table 1 Classification of organizations based upon beliefs about performance measurement (dummy data)

y xi, xj

Private xi

Private xi

Public xj

Public Private Partnership xi

Public xj

Public Private Partnership xi

Public xj

Problems in Identifying Public and Private Organizations 41

it from a theoretical prior that has been given in the older property-rights-esque literature. For this reason, I would suggest that an approach that would attempt to distinguish these paradigms should do so within the vein of Frederickson (1980) and Frederickson and Smith (2003). So one should seek to align their measurement with a given theory, but they should not expect too much from the theory. Following from the Clarke and Primo (2012) piece, one must accept that models and theories, for as elegant as they may be are simply maps.

Advantages and Disadvantages

Naïve Bayes has several advantages over other classification methods. For one thing, it is incredibly simple—essentially it is just counts and dividing. Additionally, even if the assumptions, i.e. independence, it still performs surprisingly well and the lack of independence can be corrected by weighting schemes. Additionally, if one wishes to classify organizations as public or private but does not have a lot of data then naïve Bayes is a good bet since other maximum likelihood functions will tend to over-fit with low N data.

In discussing classifying organizations as public or private, one is essentially discussing a probabilistic response based upon a series of characteristics with high dimensionality. The value of naïve Bayes over methods such as dual factor classifications (quadrant classification) in this regard is that it can account for multiple factors thus allowing a researcher to include more factors in their classification. As such one is essentially creating probability distributions based upon the characteristics of an organization. This has both advantages and disadvantages. The advantages, outside of ease of implementation, is that the method is robust to different distributions of characteristics. It is possible to combine the probability distributions from the classification with utility functions for particular characteristics. So in the case of Miller and Moulton (2013) the utility of a private organization providing a public good changed based upon the density of public service providers in the area, i.e. the utility to the private provider of offering the public service changed. With naïve Bayesian it possible to model with based upon the p(y|x) since even if the utility function changes, p(y|x) does not (Murphy 2006).

A disadvantage, which can be easily overcome, is the problem of zero-conditional probability. If a given class of characteristics never occur together then the frequency- based probability will be zero. This is a problem since it essentially wipes out all other information in those probabilities when they are multiplied in the p(y|xi)=p(y|x1)* p( y|x2)*…*p(y|xn) manner. A correction for this is Laplace smoothing which is just adding one to each count. So if for example the original probabilities were based on 0/1000; 650/1000; and 350/1000 the probabilities would be 0, 65 % and 35 %, but when we multiplied these out we would get 0. With Laplace smoothing they would be 1/1003; 651/1003; and 351/1003, i.e. 0.0009 %; 34.995 %; and 64.905 %. The “corrected” probabilities are close to the “uncorrected” originals and the zero-probability problem is avoided.

With that in mind, I do not offer any specific advice on how to measure the categories {M,I,Q,V,O,K,F} since I am espousing no particular theory. As stated before, my goal here is demonstration with an eye towards theories and empirics as opposed to espousing any particular theory of public administration.

42 J. Rauh

Data Development

I developed a data set based upon the premises outlined above. I created 1,000 observations of three different types of organizations: public, private and public- private. For all categories except O these an organization may take on any point between 0 and 1. For purposes of testing I simply generated a random set of data with this parameter, so M,I,Q,V,K&F each range between 0 and 1 for any given point. For CPr the value for O can range between 0.9 and 0.5; for CPu the values can range between 0.1 and 0.5 and for CPP the values can range between 0.1 and 0.9. I realize that this is an imposed constraint but I imply it merely for the purposes of demonstration. Otherwise organizations all organizations would range between 0 and 1 for all categories and would ostensibly be random. My method seems to make sense though given that I distinguish control from ownership. Rainey et al. (1976) identifies ownership as the distinguishing feature of public and private. However because I allow the other categories to range between 0 and 1 I satisfy the requirements set forth by Moulton (2009) that while ownership is a key feature, distinguishing public versus private must allow for other factors such as control, managerial motivation, etc. See Table 1 for description. The result is that we see organizations with distributions within categories that are roughly the same given the N=1,000. What this means is that on average an organization can take any set of values on any dimension of publicness with the exception of ownership which is the constrained dimension Table 2.

For the naïve Bayes the posterior distribution of CPr is given as:

p CPrð Þ ¼ P CPrð Þp M

� � �CPr

� � p I

� � �CPr

� � p Q

� � �CPr

� � p V

� � �CPr

� � p O

� � �CPr

� � p K

� � �CPr

� � p F

� � �CPr

� �

p CPr; CPu; CPPf g M; I; Q; V; O; K; Ff g Thr initial distribution is: p(CPr)=0.33; p(CPu)=0.33; p(CPP)=0.33 Table 3.

Results

Table 4 since the posterior distribution is greater in the case of an organization ranging between ownership being public and private we would predict that that the generated dataset represents a public-private relation. Of course the actual representation is evenly split between Public (0:33 ), Private (0:33 ) and Public-Private (0:33 ). This has two initial implications: from a measurement standpoint Moulton (2009) and Bozeman and Moulton (2011) are correct in their arguments that publicness is more than just

Table 2 Average for each category within each type (σ2)

M I Q V O ρOPu:OPp= 0.010&ρOPr: OPp=−0.020

K F

CPr 0.506 (0.086) 0.502 (0.081) 0.488 (0.087) 0.508 (0.088) 0.706 (0.019) 0.508 (0.084) 0.475 (0.073)

CPu 0.505 (0.082) 0.472 (0.083) 0.517 (0.080) 0.502 (0.085) 0.305 (0.019) 0.517 (0.083) 0.512 (0.082)

CPP 0.492 (0.088) 0.513 (0.087) 0.527 (0.087) 0.493 (0.077) 0.481 (0.062) 0.497 (0.081) 0.477 (0.085)

Problems in Identifying Public and Private Organizations 43

ownership. If we only constrain ownership to a formal definition of at least half publicly owned or at least half privately owned then we automatically allow that to be the determinate factor. In other words, any set of organizations which does not conform to that definition will not be distinguishable as either public or private but something in between. Thus it is necessary that we rely on the other categories in order to determine public versus private. Because of this though, there must also be non- randomness in the other categories which are also associated with public versus private organizations. This goes beyond classifying organizations as public or private or something in between based upon single factors of publicness. Rather it emphasizes that the other factors must work together in a manner that has at least some degree of non-randomness to it.

It is not enough to look at ownership as an indicator of publicness and then measure organizational activity on a public service outcome (Antonsen and Jorgensen 1997; Anderson 2012). Even private organizations can range in their offerings of services that are beneficial to the public—perhaps dictated by their feelings of corporate social responsibility (McWilliams and Siegel 2001). A wide range of public service outcomes are available within any given institution. These may be tied to a private organization’s perception of social responsibility, to differences in goals (see the Chic-Fil-A example). Furthermore, the proposition that outcomes that are good for the public are dependent upon public service outcomes leaves raises a number of additional questions. What are the organizational factors responsible? Does a more privatized structure work better in some places than other? (I think the answer would be yes). What other publicness indicators are associated with good public outcomes?

We know that other dimensions of publicness are important in identifying organizations as public or private. Studies from different areas have identified the importance of different dimensions. Lubienski (2003) looks at the use of private strategies to incentivize students to perform in schools. Rauh (2011) looks at the increasingly private nature of different educational choice programs and the effects upon public outcomes. Anderson (2012) examines the role of increased autonomy in the performance of public hospitals. When it comes to identifying the factors around how public or private these organizations are given these changes there are not

Table 3 Probability density function for individual categories

M I Q V O K F

CPr 1.366 1.404 1.353 1.348 0.929 1.376 1.471

CPu 1.404 1.381 1.409 1.367 0.001 1.385 1.386

CPP 1.347 1.349 1.351 1.435 6.199 4.881 1.353

Table 4 Predictions from Naïve Bayesian categorization

Classification Prediction

Private Organization 4.362 %

Public Organization 0.005 %

Ranging from Public to Private 95.634 %

44 J. Rauh

definitive answers. I would argue that the findings from the classifications above serve as a good indication of why there are not good answers.

If we want to find means of adequately defining organizations as public or private then there needs to be a consideration of public or private along all dimensions. It is easy to say that organizations have some public features and some private features and at times they will display more of one and less of the other. But how far must an organization go in one direction or another before we can say, “this organization has taken on so many dimensions of a private organization that we can classify it as private;” or “this organization has taken on so many features of a public organization that we can classify it as public?” So while it is true that the publicness literature has called for a mutli-dimensional approach, much public administration scholarship continues to equate public organizations with those that are publicly controlled—a point which Moulton (2009) and Bozeman and Moulton (2011) obviously take issue with. However allowing that dimensions of publicness exist along a spectrum also allows that combinations of those dimensions must be arranged in some minimum way if we are to say that organization X is public and organization Y is not. This means that classification schemes must expand to consider how an organization fits into all dimensions of publicness not just those that are most relevant to public outcomes.

References

Alchian, A. A. (1977). Some economics on property rights. In A. A. Alchian (Ed.), Economic forces at work. Indianapolis: Liberty Fund.

Anderson, S. (2012). Public, private, neither? Publicness theory and the analysis of healthcare organizations. Social Science and Medicine, 74(3), 313–322.

Andrews, R. W., Boyne, G. A., & Walker, R. M. (2011). Dimensions of publicness and organzational performance: a review of the evidence. Journal of Public Administration Research and Theory, 21(s3), i301–i319.

Antonsen, M., & Jorgensen, T. B. (1997). The publicness of public organizations. Public Administration, 75(2), 337–357.

Arnold, R. D. (1990). The logic of congressional action. New Haven: Yale University Press. Babbie, E. (2005). The basics of social research. Independence: Cengage. Bozeman, B. (1987). All organizations are public: Bridging public and private organization theory. San

Francisco: Jossey-Bass. Bozeman, B. (2007). Public values and public interest: Counterbalancing economic individualism.

Washington DC: Georgetown University Press. Bozeman, B. (2013). What organization theorists and public policy researchers can learn from one another:

publicness theory as a case-in-point. Organization Studies, 34(2), 169–188. Bozeman, B., & Bretschneider, S. (1994). Understanding Red Tape and Bureaucratic Delays. In A. Halachmi

& G. Bouckaert (Eds.), The enduring challenges in public management. San Francisco: Jossey-Bass. Bozeman, B., & Moulton, S. (2011). Integrative publicness: a framework for public management strategy and

performance. Journal of Public Administration Research and Theory, 21(suppl 3), i363–i380. Clark, C. W. (1973). Profit maximization and the extinction of animal species. Journal of Political Economy,

81, 950–961. Clarke, K., & Primo, D. M. (2012). A model discipline political science and the logic of representations. New

York: Oxford University Press. Dahl, R. A., & Lindblom, C. E. (1953). Politics, economics and welfare. New Brunswick: Transaction

Publishers. DeHart-Davis, L., & Kingsley, G. (2005). Managerial perceptions of privatization: evidence from a state

department of transportation. State and Local Government Review, 37(3), 228–241. Emmert, M., & Crow, M. (1988). Public, private and hybrid organizations: an empirical examination of the

role of publicness. Administration and Society, 20(2), 216–244.

Problems in Identifying Public and Private Organizations 45

Finlay, W., Martin, J. K., Roman, P. M., Blum, T. C. (1995). Organizational structure and job satisfaction: do bureaucratic organizations produce more satisfied employees? Administration & Society, 27(5), 427.

Flinders, M., & John, P. (2013). The future of political science. Political Studies Review, 11(2), 222–227. Frederickson, H. G. (1980). New public administration. Tuscaloosa: University of Alabama Press. Frederickson, H. G., & Smith, K. (2003). The public administration theory primer. Boulder: Westview Press. Frumkin, P., & Galaskiewicz, J. (2004). Institutional isomorphism and public sector organizations. Journal of

Public Administration Research and Theory, 14(3), 283–307. Gaskell, J., & Levin, B. (2012). Making a difference in urban schools: Politics and pedagogy. Toronto:

University of Toronto Press. Gill, B. P., Timpane, P. M., Ross, K. E., & Brewer, D. J. (2001). Rhetoric versus reality: What we know and

what we need to know about vouchers and charter schools. Pittsburgh: RAND. Hand, D. J. & Yu, K. (2001). Idiot’s Bayes: not so stupid after all. International Statistical Review, 69(3), 385–398. Heinrich, C. J., & Fournier, E. (2004). Dimensions of publicness and performance in substance abuse

treatment organizations. Journal of Policy Analysis and Management, 23(1), 49–70. Jones, B. D., & Baumgartner, F. R. (2005). A model of choice for public policy. Journal of Public

Administration Research and Theory, 15(3), 325–351. Jones, B. D., & Baumgartner, F. R. (2012). From there to here: punctuated equilibrium to the general

punctuation thesis to a theory of government information processing. Policy Studies Journal, 40(1), 1–19. Lewis, M. W., & Grimes, A. J. (1999). Metatriangulation: building theory for multiple paradigms. Academy of

Management Review, 24(4), 672–690. Lubienski, C. (2003). Istrumentalist perspectives on the ‘public’ in public education: incentives and purposes.

Educational Policy, 17(4), 478–502. McWilliams, A., & Siegel, D. (2001). Corporate social responsibility and financial performance: correlation or

misspecification? Strategic Management Journal, 21(5), 603–609. Meier, K. J., & O’Toole, L. J. (2011). Comparing public and private management: theoretical expectations.

Journal of Public Administration Research and Theory, 21(Supplement 3), i283–i301. Meyer, M. W., & Williams, R. O. (1977). Comparison of innovation in public and private Sectors: An

exploratory study. Washington, D C: National Science Foundation. Miller, S. M., & Moulton, S. (2013). “Publicness in policy environments: a multilevel analysis of substance

abuse treatment services,” Journal of Public Administration Research and Theory. Available at: http:// jpart.oxfordjournals.org/content/early/2013/01/10/jopart.mus065.short.

Miller, G. J., & Yang, K. (2007). Handbook of research methods in public administration (2nd ed.). Boca Raton: Taylor and Francis Group.

Morgan, G. (1980). Paradigms, metaphors and puzzle solving in organizational theory. Administrative Science Quarterly, 25(4), 605–622.

Moulton, S. (2009). Putting together the publicness puzzle: a framework for realized publicness. Public Administration Review, 69(5), 889–900.

Moynihan, D. P. (2009). Through a glass darkly: understanding the effects of performance regimes. Public Performance Management Review, 32(4), 586–598.

Moynihan, D. P., & Pandey, S. K. (2010). The big question for performance management: why do managers use performance information? Journal of Public Administration Research and Theory, 20(4), 849–866.

Murphy, K. (2006). Naive Bayesian classifiers. Available at: http://www.google.com/url?sa=t&rct=j&q= &esrc=s&frm=1&source=web&cd=1&ved=0CCoQFjAA&url=http%3A%2F%2Fwww.cs.ubc.ca% 2 F ~ m u r p h y k % 2 F Te a c h i n g % 2 F C S 3 4 0 - F a l l 0 6 % 2 F r e a d i n g % 2 F N B . p d f & e i = 1TZpUva6O6fj2AWsp4HoCQ&usg=AFQjCNGdA77BO-d_B3Vy9BEL6dGTU0WEvQ&sig2= WfZEqueIpsqCfm6OMNDwMQ.

Perry, J. L. (2000). Bringing society in: toward a theory of public-service motivation. Journal of Public Administration Research and Theory, 10, 471–488.

Perry, J. L., & Rainey, H. (1988). The public-private distinction in organization theory: a critique and research strategy. The Academy of Management Review, 13(2), 182–201.

Perry, J. L., & Wise, L. R. (1990). The motivational bases of public service. Public Administration Review, 50(3), 367–373.

Perry, J. L., Brudney, J. L., Coursey, D., & Littlepage, L. C. (2008). What drives morally committed citizens? A study of the antecedents of public service motivation. Public Administration Review, 68(3), 445–458.

Rainey, H. G. (2009). Understanding and managing public organizations (4th ed.). San Francisco: Wiley/ Jossey-Bass.

Rainey, H. G., Backoff, R. W., & Levine, C. H. (1976). Comparing public and private organizations. Public Administration Review, 36(2), 233–246.

46 J. Rauh

Ratanamahatana, C. A., & Gunopulos, D. (2002). Scaling up the naive Bayesian classifier: Using decision trees for feature selection. In Proceedings of Workshop on Data Cleaning and Preprocessing (DCAP 2002), at IEEE International Conference on Data Mining (ICDM 2002). Maebashi, Japan.

Rauh, W. J. (2011). “The utility of online choice options: do purely online schools increase the value to students? Education Policy Analysis Archives 19(34). Retrieved from: http://epaa.asu.edu/ojs/article/view/923/945.

Rish, I. (2001). An empirical stdy of the naive Bayes classifier. Available at: http://www.cc.gatech.edu/home/ isbell/classes/reading/papers/Rish.pdf.

Stout, L. (2002). Bad and not-so-bad arguments for shareholder primacy”, 75 S. California Law Review, 1189(75).

Tompkins, M., & Jos, P. (2009). Keeping it public: defending public service values in a customer service age. Public Administration Review, 69(6), 1077–1086.

Train, K. (2003). Discrete choice methods with simulation. New York: Cambridge University Press. Warner, M. E., & Hefetz, A. (2010). “Service characterization and contracting: The importance of citizen

interest and competition”, In The municipal year handbook 2010 (pp. 19–27). Washington DC: International City County Management Association.

Williamson, O. E. (1999). Public and private bureaucracies: a transaction cost economics perspective. Journal of Law and Economic Perspectives, 15(1), 306–342.

Wolf, C. (1988). The rise of market forces. Santa Monica: RAND Corporation.

Jonathan Rauh is a PhD candidate at the University of South Carolina. His research interests include accountability, organizational theory, and healthcare and education policy. His work has appeared in Administration and Society, Education Policy Analysis Archives, and the Journal of Political Science Education.

Problems in Identifying Public and Private Organizations 47

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.