 |
 |
|
||||||||
|
|
|
|
|
|
|
||
|
|
|
||||||
|
|
|
Volume 1: No. 4, October 2004
ORIGINAL RESEARCH
|
||||||||
|
|
|
|
Glossary For the discussion in this article, these nontechnical network analysis definitions may be useful (28): Connectedness: A network graph is connected if there is a path or tie between every pair of actors in the graph. Therefore, all pairs of actors are reachable in a connected graph. Density is defined here as the proportion of possible lines or ties that are actually present in a network graph. Because it represents a proportion, density ranges from 0 to 1. Centrality provides a measure of how central an actor is within a network. Actors that are highly central are interpreted as controlling the flow of information or resources within the network. Betweenness is a measure of centrality based on how often an actor in a network is found in the shortest pathway between other actors in the network. The equation for normalized betweenness can be found in Appendix B. It is used in this study to evaluate frequency of contact and control over network communication. Higher scores indicate higher control. Prestige is a value that is commonly examined in a directional relationship as a way to identify prominent actors in a network. A prestigious or prominent actor is one that is the object or recipient of many ties in the network. Normalized indegree is used to measure prestige; it indicates the number of directional ties terminating at or pointing toward an actor. A higher indegree score indicates higher prestige. |
|
|
|
|
|
|
Contact frequency
One important use of social network analysis is to identify the most
important actors in a network, which are considered to be in strategic locations
within the network (28). Our measure of contact frequency is nondirectional,
meaning that the relation between two actors in the network holds simultaneously for both actors. So, if Agency A has contact with Agency B at least monthly, then
Agency B must also have contact with Agency A. One way to determine the
important actors with this type of relation is to examine centrality using the
measure of betweenness. We examined the normalized betweenness of each actor
(30).
Flow of money
Exchange of money among network partners was a directional relation, meaning
that the tie between two actors has an origin and a destination. In this case,
Agency A may send money to Agency B, but not necessarily vice versa. For this
construct, we used Taylor抯 influence to measure the amount of financial
influence of one agency over another (31). (Appendix B provides more information
on Taylor抯 influence.) A negative influence value indicates that an agency
had a preponderance of receiving over sending money to other agencies in the
network, a positive value indicates a preponderance of sending over receiving, and a neutral value indicates a balance of sending and
receiving.
Relationship productivity
Our productivity variable was also a directional relation. Agency A may feel
they have a productive relationship with Agency B, but the feeling may not be
mutual. A common way of identifying prominent actors in a directional relation
is by examining prestige, which we measured using normalized indegree (28). (See
Appendix B for definition and equation.) In our study, a prominent or
prestigious agency would be one with whom many others felt they had highly
productive relationships. A higher indegree score indicates higher prestige.
Group-level indices
In addition to the actor-level measures described above, we also examined
some group-level indices to facilitate comparisons across networks and states. A
group betweenness centralization index was calculated for each contact frequency
network, which indicates variability of the betweenness of members of the
network. A high betweenness centralization score indicates a hierarchical
network structure, where it is more likely that a single agency in the network
is quite central, while remaining agencies are less central. We did not examine
a group-level index for prestige because little research has been done to
adequately develop and validate network-level prestige indices (28). We also
examined the density of each network for contact frequency, relationship
productivity, and money flow. Density indicates the proportion of all possible
ties that are actually present in the network. The above constructs were also
compared with other state network variables, such as funding level, receipt of
the Robert Wood Johnson Foundation (RWJF) Smokeless States grants, and location of
partner agencies.
Qualitative network constructs were collected via open-ended questions about the perceived effectiveness of the state抯 tobacco control network and suggestions for methods to improve the network抯 effectiveness. Each interview was transcribed verbatim and imported into the qualitative data management software NUD*IST (QSR International Pty Ltd, Melbourne, Australia). Each transcript was then coded using a detailed codebook developed during the pilot test by two trained staff members. Inter-rater reliability for coding was 83.7%. The coded text units were entered into NUD*IST, and a report was generated for each construct (e.g., network, financial climate). Analysis teams consisting of two trained staff members independently analyzed the reports to identify major themes or ideas. The team then met to discuss the results and arrive at consensus on major themes.
The five state tobacco control programs represented a variety of funding levels, network sizes, and geographic locations (Table 1). Annual funding levels for entire state programs ranged from $4.2 million (Wyoming, fiscal year 2003) to $52.3 million (New York, fiscal year 2003). According to the CDC抯 minimum recommendations for state tobacco control funding (1), Michigan had low funding (10% of CDC抯 recommendation), New York, Wyoming and Washington had middle funding levels (55%, 57%, and 62%), and Indiana had a high funding level (97%). Three of the statewide coalitions received a Smokeless States grant from the RWJF at the time of the evaluation. The average number of partner agencies per state network was 14, and the number of cities where partner agencies were located ranged from a few (two in Indiana) to as many as 11 (New York).
Four of the tobacco control networks were led by the tobacco control program within the state抯 department of health. The exception was Indiana抯 program, which was led by an independent tobacco control agency. Lead agencies usually included as key partners in the network were voluntary agencies, advocacy agencies, statewide and regional coalitions, and contractors with the lead agency. Partner agencies unique to some states included government agencies, political figures, trust fund agencies, media firms, and funding agencies. (Appendix C lists agencies, along with their abbreviations, included in each state network.)
Figure 1 depicts the monthly contact network for each of the five state programs. The amount of control over communication flow each agency had relative to other agencies in the network is based on centrality betweenness scores presented in Table 2 (the higher the score, the more control over communication flow in the network).
Figure 1. Monthly contact network for five state tobacco control programs. A line connects two agencies that had contact with each other at least once a month via meetings, phone calls, or e-mails. Colored dots represent the amount of control over communication flow each agency had relative to other agencies in the network as determined by scores for betweenness. All acronyms are spelled out in Appendix C. (This information is also available in a tabular format.)
Two of the states presented contrasting patterns of network communication structure: New York and Michigan. New York is an example of a very hierarchical structure because of its high centralization index (40.8%), high mean betweenness score (5.1), and standard deviation (11.0). Most agencies in the New York network had frequent contact with one central agency (the lead agency New York State Department of Health Tobacco Control Program), which also had the most control over communication flow within the network, indicated by its high betweenness score (43.1). Betweenness scores dropped substantially for other partner agencies. The density of New York抯 network was the lowest of the five states (0.39). Figure 1 shows that many agencies had very little contact with other agencies. An agency would likely have to go through the lead agency to communicate with another agency. From qualitative analysis, we found that network partners believed the network was improving and had potential, but thought it still needed to improve its coordination and communication.
Michigan, on the other hand, had a much flatter, or nonhierarchical, communication structure. This structure is reflected by its low centralization index (10.4%), and the low mean betweenness score (3.5) and standard deviation (5.2). Partner agencies had frequent contact with many other agencies in the network. Three agencies, including the lead agency Michigan Department of Community Health Tobacco Section, the statewide coalition Tobacco Free Michigan Action Coalition (TFMAC), and the contractor Gerontology, had higher control over communication flow than other agencies in the network. Although these three agencies had higher relative betweennness scores within their own network, their scores were not very high compared to agencies in other state networks. Michigan抯 density (0.58) was higher than that of the other four states. In contrast to New York抯 contact graph (Figure 1), Michigan抯 figure shows a much more connected communication network. There were multiple routes for information flow from agency to agency. Michigan抯 partner agencies felt their network was very effective and had improved over the last few years. They believed TFMAC and a Smokefree Regulations Task Force, a group of partners focusing on clean indoor air efforts, were important network components that facilitated the process of bringing stakeholders together. Partners felt this helped strengthen the network抯 ability to work more efficiently. Yet partners thought communication could be improved by openly sharing agencies?priorities and activities.
In examining all five states, a few patterns emerged. The lead agency usually had the highest control over communication flow, which intuitively makes sense. However, statewide coalitions, which are known to bring stakeholders together on tobacco issues, usually did not have the highest communication control, except in Michigan. Most network communication structures were relatively hierarchical, except Michigan, which had a much flatter structure. The average density of the contact networks was 0.47. The densities ranged from 0.39 in New York to 0.58 in Michigan.
Figure 2 shows the money flow network for each state. Table 3 presents the financial influence scores for each agency in each network. In most states, the lead agency had the highest financial influence over the network, which corresponds with their duties as fiscal oversight agencies. The exception was Wyoming, where the legislature was included as a partner in the network. Due to the legislature抯 role in appropriating funding, it had the highest financial influence. Statewide coalitions were usually highly influenced by others in the network, except in Indiana. The statewide coalition Tobacco Smart Indiana did not send or receive money from anyone in the network at the time of the study. At the time of the evaluation, the coalition was in transition because of a loss of funding.
Figure 2. Money flow networks for five state tobacco control programs. Arrows indicate direction of money flow, and colored dots represent the relative amount of financial influence each agency had over the rest of the network. All acronyms are spelled out in Appendix C. (This information is also available in a tabular format.)
Many of the states with greater funding levels had more complex money flow networks, such as Indiana, New York, and Washington. In these states, money flowed not only from the lead agency to other agencies but also flowed among contractors and coalitions. Wyoming, which had a low funding level, had the simplest money flow network, with money mainly flowing from the legislature to the lead agency and then to contractors or coalitions. Some money was also given by other agencies to the statewide coalition. Network density related to money flow varied little among the states. The average density was 0.09.
Figure 3 depicts the productivity of relationships in the five state networks. Table 4 presents the prestige score (normalized indegree) for each agency in each network.
Figure 3. Productivity of relationships in the five state networks. An arrow from A to B indicates that Agency A felt it had a very productive relationship with Agency B. A bidirectional arrow indicates that both agencies agreed that their relationship was very productive. Colored dots represent the prestige each agency had relative to other agencies in the network. All acronyms are spelled out in Appendix C. (This information is also available in a tabular format.)
In Washington, the lead agency, Washington State Department of Health Tobacco Prevention and Control Program (WA DOH), had the highest prestige (normalized indegree 69.2), followed by the American Lung Association-Washington State Branch and the Washington Office of the Attorney General (both normalized indegree 46.2). The WA DOH was highly regarded by partner agencies. Furthermore, partner agencies were pleased that the state secretary of health had made tobacco control her highest priority. The state attorney general had been one of the lead negotiators in the Master Settlement Agreement and was overwhelmingly identified as a tobacco control champion. The statewide coalition Washington Alliance for Tobacco Control and Children抯 Health (WATCH) ranked in the middle of other Washington agencies in prestige (normalized indegree 30.8). At the time of the evaluation, WATCH was going through a transition because of a recent loss of funding and was reevaluating its role in the tobacco control network. Many partners were uncertain of the coalition抯 existence or future plans. Washington抯 productive relationships network had a low density compared to the other states (0.27).
Wyoming抯 statewide coalition, Wyoming Tobacco Use Prevention (WY TUP), had the greatest prestige score in its state (normalized indegree 81.8). The lead agency, Wyoming Department of Health Substance Abuse Division, Tobacco Prevention and Control Program (DOH TPCP), on the other hand, had a much lower score (normalized indegree 45.5). Again, these results were supported qualitatively. Partners were very pleased with WY TUP and its accomplishments. However, they believed the DOH TPCP, which houses the tobacco control program, did not provide enough support for tobacco control. In Wyoming, the Legislature had the lowest prestige score (normalized indegree 9.1) due to sentiment that the legislature as a whole was unsupportive of tobacco control. Partners believed that only a few legislators made tobacco control a priority. The Department of Education and Department of Maternal and Child Health in Wyoming also received relatively low scores in prestige. Partners believed those agencies were not as engaged in tobacco control as they could be.
Although problems may be encountered with the lead agency in some states, it usually had a very high number of productive relationships (if not the highest) across states. Statewide coalitions funded by the RWJF (New York, Wyoming, and Michigan) had higher prestige scores than coalitions that did not receive RWJF funding and were in transition (Washington and Indiana). The average prestige score for RWJF-funded coalitions was 69.9, compared to 29.7 for non-RWJF杅unded coalitions. Agencies with narrowly defined roles, such as contractors or agencies with a local focus, had fewer highly productive relationships, while state level agencies usually had higher numbers of productive relationships. The average density of the productivity networks was 0.34. The results suggested a relationship between contact network density and productivity network density. Michigan had the highest productivity density (0.41) and also had the highest contact density (0.58). New York (0.25) and Washington (0.27) had lower productivity densities and also had lower contact densities (New York, 0.39 and Washington, 0.44).
This paper presents a new construct that can be used to examine state tobacco programs: network structure. Using social network analysis to examine five state tobacco control programs led to some important observations. Key actors in the networks were highlighted, such as the lead agency of the programs. In all five states, lead agencies had high control over communication flow, many highly productive relationships, and much financial influence over the networks. The financial stability of statewide coalitions also influenced network structure. Statewide coalitions had many highly productive relationships in some states, but not as many in other states. This difference could be explained by the funding status of the coalitions. Those with RWJF funding scored higher in productivity than coalitions without RWJF funding. Funding is necessary for sustaining a stable statewide coalition and for building and maintaining high-quality relationships with others in the state.
Some patterns between network structure and basic descriptive network characteristics also emerged. For example, results suggested that geographic dispersion of a network could play a role in communication among agencies. Densities of contact networks appeared to be higher for states with partners in fewer locations and vice versa. New York had the greatest number of partner agency locations (11) and the lowest density (0.39). Conversely, Indiana had a low number of partner agency locations (2) and a high contact density (0.50). This does not suggest that tobacco control efforts should be concentrated in only a few areas of the state. In fact, it is important to have partners in many areas of the state to reach citizens with messages of prevention and cessation and to create policy change. This does suggest, however, that increased efforts may be needed to facilitate communication among agencies located throughout the state.
While the results of this descriptive study cannot directly assess causality, it was clear that communication among agencies was an important factor for having productive relationships. Results suggested that states with more dense contact networks had more dense productivity networks. Michigan had the highest contact density (0.58) and the highest productive relationships density (0.41). New York had the lowest density for contact (0.39) and productive relationships (0.25). This is logical because a productive relationship can only occur when at least some contact occurs between agencies. However, frequency of contact could be a factor here. The more often communication occurs among agencies, the more they work together and the more productive they feel their relationship is. It is relatively easy to investigate the amount of contact occurring among agencies. Little contact among agencies could be a symptom of other problems in the network that may lead to lower productivity.
To further pursue this, we calculated the relationship between contact and productivity for each of the five states by determining the graph correlation using the quadratic assignment procedure (32). Table 5 presents these correlations and shows that for these states there is a significant positive correlation between contact and relationship (average correlation = .56). Therefore, we see that tobacco control agencies that have more frequent contact with each other are more likely to report highly productive relationships.
Surprisingly, a relationship between funding level and network connectedness was not suggested by our results. Both New York and Indiana had high funding dollar amounts, but Indiana抯 network seemed very well connected, while New York抯 network was less connected. Michigan had a lower funding dollar amount and had a very connected network. Therefore, funding level does not seem to be a driving force for how connected a state program can be.
Nuances in state tobacco control programs also affected the structure of the networks. For example, Wyoming抯 tobacco control program had the only lead agency that was placed under the Substance Abuse Division at the Department of Health. Partners believed the program did not provide enough support for tobacco control, which caused this lead agency to have fewer productive relationships than lead agencies in the other states. The Washington Attorney General抯 prominent role in the Master Settlement Agreement and continued support for the program made the attorney general a unique partner in the network, and partners believed their relationships with the attorney general were highly productive.
Results of this study should be interpreted with some caution. Because our sample size included only five state networks, results are not very generalizable. Furthermore, state networks presented do not include all agencies involved in the state network. A limited number of agencies could be interviewed because of the very large number of agencies involved in a state抯 program and the study team抯 limited resources. However, the type of sampling methodology employed here, which included a list of key partners identified by tobacco control program managers and the addition of some other agencies as suggested by the key partners, resulted in including the most important tobacco control agencies in the evaluated tobacco control network. Although only one individual per agency was usually interviewed, we believe those responses represent the viewpoint of the entire agency on quantitative network constructs. When multiple individuals from an agency were interviewed, their responses for quantitative constructs were highly correlated. The productivity of relationships was dichotomized as 搗ery productive?or 搉ot very productive.?nbsp; Dichotomization of this variable was necessary for analysis, and we chose to use 搗ery productive?as the cutoff to highlight the highest productivity level between agencies. Doing so could be a disadvantage because of the loss of variability in the measure. Finally, our analysis is based on reports of individuals from partner agencies, which may not be accurate, or at times were simply missing. Reported responses may differ from actual, observed interactions among partners. Possible bias in reporting or from missing data is an inherent limitation in key informant interviews.
The CDC has developed an extensive logic model for tobacco prevention in an effort to identify specific outcome indicators for comprehensive tobacco control programs (33). The emphasis in the CDC logic model is on short-, medium-, and long-term outcomes, such as smoking initiation rates, cigarette prices, and existence of tobacco control policies.
The work presented in this paper represents a first step in developing measures of process indicators as opposed to outcome indicators. That is, very little is known about the organizational structures and processes that may influence successful tobacco control outcomes. Investigating structures of state tobacco control networks can help shed light on the highly complex process of coordinating a state tobacco control program. Social network analysis provides an objective and relatively simple and inexpensive way to examine this process. One important way this type of analysis could be used is to examine the effect of state budget cuts on state tobacco control programs. Other than tracking the amount of money states spend on tobacco control, it is difficult to measure the impact of budget cuts on the process of implementing a statewide program. Network structure could be compared before and after budget cuts to determine changes in network characteristics and size. More importantly, future research could compare network analysis measures with the success of state programs in decreasing smoking rates or passing tobacco control policies to establish the link between network characteristics as process and program outcome indicators. Furthermore, this type of organizational network analysis represents a new way to perform program evaluation, in which the emphasis is less on simple counts of program activities and more on the documentation of structural and process changes that result from effective programs (26). In summary, network analysis of state tobacco control programs can identify the network抯 structure and connectedness of the agencies within the program. Once linked with program outcomes in future research, this analysis could be used to inform necessary development or organizational changes to tobacco control programs and produce better outcomes.
We wish to thank the tobacco control partners in Washington, Indiana, Wyoming, New York, and Michigan for their participation in this research. This research was supported by a grant from the American Legacy Foundation with collaboration from the CDC Foundation and scientific and technical assistance from the CDC. Results of this paper do not necessarily represent the views of the American Legacy Foundation or the CDC Foundation, their staff, or boards of directors.
Corresponding author: Nancy Mueller, MPH, Center for Tobacco Policy Research, Saint Louis University School of Public Health, 3545 Lafayette Ave, Suite 300, St. Louis, MO 63104. Telephone: 314-977-4027. Fax: 314-977-3234. E-mail: mueller@slu.edu.
Author affiliations: Melissa Krauss, MPH, Douglas Luke, PhD, Center for Tobacco Policy Research, Saint Louis University School of Public Health, St. Louis, Mo.
|
|
|
|
|
|
|
|
The opinions expressed by authors contributing to this journal do not necessarily reflect the opinions of the U.S. Department of Health and Human Services, the Public Health Service, the Centers for Disease Control and Prevention, or the authors’ affiliated institutions. Use of trade names is for identification only and does not imply endorsement by any of the groups named above.
Privacy Policy | Accessibility This page last reviewed March 30, 2012
|
|
