A Survey of Gambling Behaviors in Michigan, 2001



A Survey of Gambling Behaviors in Michigan, 2006

By

David J. Hartmann, Ph.D.

Department of Sociology

Western Michigan University

This study was conducted by the Kercher Center for Social Research at Western Michigan University for The Michigan Department of Community Health. Data collection was subcontracted to EPIC MRA Inc. of Lansing, Michigan. The views expressed are those of the author and do not necessarily reflect those of any of the above organizations.

Acknowledgments

This work is based on the surveys done in 1997, 1999, and 2001 with Arlen Gullickson and the Evaluation Center at Western Michigan University. The professionalism and high quality of the work of EPIC MRA is here gratefully acknowledged, as is the assistance of Syprose Owaja on all aspects of the research design. As always, our deepest debt is to the Michigan residents who gave generously of their time and experience to produce these results.

The survey instrument was adapted from the work of Rachel Volberg who was a consultant on the 1997 project. Her published material and her willing assistance during that first project are again gratefully acknowledged. Her work since that time provides continued guidance and perspective, and remains at the core of prevalence research.

CONTENTS

Acknowledgments i

Table of Contents ii

Introduction 1

Characteristics of the Sample 4

Results 7

Detailed Results 14

Results for Problem Gamblers 20

Summary 23

Appendix A: Survey 25

Appendix B: Counties in Geographic Regions 41

References 42

Introduction

The 2006 Survey of Gambling Behaviors in Michigan is the fourth iteration of a project begun in 1997 and primarily designed to provide an estimate of problem gambling in the state. While there were two-year intervals between that first study and replications in 1999 and 2001, it has now been five years since a statewide prevalence estimate has been produced in Michigan. While minor changes were implemented in each iteration of the earlier studies (c.f., Hartmann and Gullickson 2001), this year’s project is, as closely as possible, a replication of the 2001 research. The most important changes made up to and including the 2001 project were: 1) inclusion of questions on Internet gambling (added in 1999) and on suicidal ideation related to gambling and use of the State Problem Gambling Help Line (added in 2001), 2) the design sampled and collected responses so as to produce samples of at least size 384 from each of five regions in the State of Michigan: The City of Detroit, the remainder of the Detroit metropolitan area, East Michigan, West Michigan, and the Upper Peninsula (see Appendix B). In 1999 the Detroit Metropolitan Region included the City and in 1997, Wayne County rather than the City was estimated separately from the rest of the Metropolitan Area. This year, 400 responses were obtained in each of the five regions.

The current design allows inference of the rate of problem gambling within each region with a reasonable degree of precision based on sampling error (plus or minus 3 percentage points[1]) and allows combination of those regions in proportion to their contribution to the adult population of the state in an aggregate data set. As described more fully in the section called “Characteristics of the Sample,” this year’s state aggregate data set contains 957 interviews and therefore has precision based on sampling error of plus or minus 1.9 percentage points[2] for the rate of problem gambling.

A consistent challenge of prevalence studies is that, since the rate of problem gambling is low, regional and even statewide samples yield a small number of persons scoring with a problem on the South Oaks Gambling Screen (SOGS: discussed below). For obvious reasons, including the planning of helping strategies, this is an important population to sample and about which to make inference. In both 2001 and this year, additional interviewing was therefore done to increase the number of respondents scoring as a “problem” or “probable pathological gambler” on the SOGS to 200 thereby allowing more precise analysis of this important subgroup. In 2001, we used a special sample of persons with an expressed interest in gambling as a form of recreation to efficiently increase the number of responding problem or probable pathological gamblers. The non-comparability of this targeted sample with the random adult sample used for the main study made combining the problem and probable pathological gamblers from the two datasets problematic. To address this concern, this year additional sampling to obtain the needed numbers was done using the same population of adult residents of the state as was used for the main study. While this is less efficient, all 206 problem and probable pathological gamblers interviewed are now drawn from and represent the adult statewide population. Fully 118 interviews completed from this additional interviewing were added to the 88 interviews obtained from the original statewide calling.

As before, the primary aim of the survey is to establish a precise estimate of problem gambling in the population of Michigan residents 18 years and older. The 1997 study was required to establish this estimate with precision due to sampling error of no more than plus or minus 1 point (Gullickson and Hartmann 1997). This led to a design through which 3,942 responses were completed. Subsequent iterations were allowed to produce statewide estimates with slightly larger confidence bands and so allowed substantial data collection savings. The statewide samples were of size 1,211 in 2001 and 957 in 2006.

The standard in prevalence studies, including our earlier work in Michigan, is to administer the survey through a Computer Assisted Telephone Interviewing (CATI) approach utilizing a random-digit dial (RDD) telephone sample. The CATI system automates and documents the distribution of numbers to interviewers while also recording the disposition of each call and storing completed interviews in a database. Efficiency of administration is enhanced through automated advancing through contingency branches in the survey and data entry errors are minimized through range restrictions and similar verification checks. In our survey, for example, the SOGS score must be calculated to determine whether the section of the survey for problem gamblers should be completed. This would be very difficult to accomplish in a non-computerized format. The random digit numbers themselves are obtainable in a variety of ways but generally trade off inclusiveness for efficiency. For example, area codes and three digit prefixes are typically the starting point for randomizing the last four numbers while even potential subsets of these suffixes are systematically vetted to increase efficiency of hit rates (actually connecting to a residential number).

As documented in our earlier reports and most fully in Gullickson and Hartmann 1997, the original form of the survey instrument used in 1997 was adapted from Rachel Volberg’s survey of New York State in 1996 (Volberg 1996c) and uses the South Oaks Gambling Screen (SOGS) as the basis for estimates of problem gambling. Again, only minor changes to the instrument have been made in our subsequent studies.

Since the South Oaks Gambling Screen (Lesieur & Blume, 1987) is the basis for the prevalence estimates made in this study, the brief description offered in earlier reports is repeated here. The SOGS asks about a range of behaviors and orientations toward gambling and is highly correlated with the APA’s DSM-III-R (Diagnostic and Statistical Manual of Mental Disorders, 3rd ed.-revised) criteria for pathological gambling (American Psychiatric Association, 1987). It has possible scores of 0 to 20 with 0 through 2 considered nonproblem gambling, 3 through 4 identified as “problem gambling”, and 5 or more identified as “probable pathological gambling.” As is customary in current use of the SOGS, we asked each of the 20 scored questions for two time frames, “ever” and “in the past year.” These give rise to a “lifetime” SOGS score and a “current” (past year) SOGS score. Since a person must score a point on the lifetime question to be asked the past year question, the lifetime score is the basis for admission to the problem gamblers section of the survey.

Other scales, based on the DSM-IV for example, are available. In fact, Volberg herself has regularly predicted that these other scales might come to supplant the SOGS but, as of her summary of the field in 2004 (Volberg 2004), that had not yet happened. So, because so large a preponderance of geographic prevalence studies (including studies at the state and Canadian Province level) used the SOGS, including the 1997, 1999, and 2001 Michigan studies, we retain it here.

Characteristics of the Sample

EPIC/MRA reports that in the geographic sampling used to produce the regional and statewide estimates, a total of 2,000 responses were obtained with a refusal rate of just over 71%. This rate for the main study is higher than the 65% rate obtained in 2001 but is well within the expected range for telephone surveys over the past five years. In fact, a recent study in British Columbia reported a similar 73% refusal rate and also pointed out that a review of national omnibus surveys showed an average refusal rate of 77% (British Columbia 2003). Groves et al (2004) report that that even the Behavioral Risk Factor Surveillance System, one of the best funded and prestigious telephone based household surveys, showed an increase in median

non-response[3] across states from about 30% in 1991 to almost 50% by 2001 (p. 187). Since most statewide gambling prevalence studies were done some time ago, their refusal rates are a bit lower. The last two state-wide surveys we reviewed in 2001 showed a 64 percent rate in New York in 1996 and a 60 percent rate for Louisiana in 1995 at about the time we achieved the 65% rate in Michigan. Note that our earlier studies also had somewhat better refusal rates: 57 percent rate in 1997 and 55 percent in 1999. It is reassuring but not sufficient that several studies (British Columbia 2003, Volberg 2004) point out that the quality of prevalence rate estimates seem to be robust with respect to refusal rates.

Since samples sizes of 400 were collected by region to allow inference at acceptable levels for each part of the state, a representative statewide sample could not be a simple aggregation of the regions. A weighting procedure was used to produce a statewide sample of size 957 that is weighted to represent the adult population of Michigan at the county level. This resulted in an error band at a 95% confidence level for problem gambling rates with a precision of plus or minus 1.9 points based on sampling error (the band was plus or minus 1.7 points in 2001). The weighted sample includes 403 cases from the Metropolitan Detroit sample, 179 from the east counties, 251 from the west, 35 from the U.P., and 89 from the City of Detroit. That weighted sample is used throughout this report as the “state sample.”

Sampling variation due to sample size is only one source of error in inference. The real concern is response bias and a standard check on this, particularly in the presence of high refusal rates, is to directly compare the obtained demographic characteristics of the sample against other estimates of those population characteristics in which one has some confidence. Table 1 does this for the statewide weighted sample against the 2000 Census figures for the state.

In reviewing this comparison, it is important to note that telephone surveys are used for prevalence studies because, despite reduced response rates, they can produce an efficient tradeoff of cost and response bias. Random digit dial (RDD) approaches, in particular, are preferred because they address the most obvious sources of bias in telephone sampling, access to unlisted numbers. Nevertheless, the RDD telephone survey has known weaknesses. First, most survey organizations exclude cell phones. There is an increasing percentage of the population that does not have a landline (perhaps as high as 5-10%). Also, as discussed in earlier reports, telephone surveys in general often under-represent males, poor people, and younger respondents and therefore tend to under-represent characteristics associated with male sex, low income, and youth. Several factors are likely to be in play. First, men are less likely to answer the phone when a woman also resides there. Second, the poor simply are less likely to own a phone. Third, participation rates in survey research are directly related to education. Furthermore, poorer families and young householders may be less likely to have an adult at home in the evening when the bulk of contact attempts are made (due to one adult households and late shift work). Poor households also tend to have a younger age structure, which is also related to presence in the home and willingness to participate. In any event, most telephone surveys expect to under-represent men, the young, the poor, and the less educated and consequently black and central-city residents as well.

Each recent statewide gambling study we reviewed reported these biases, especially with regard to education and income. A standard correction for each response rate variation is to weight the underrepresented category for analyses. Most of the statewide gambling studies did not do this, however. In her Iowa report, Volberg contends that, “To maintain comparability with results from the 1989 survey from Iowa, as well as with results from surveys in other United States jurisdictions, it was deemed advisable to caution readers regarding these prevalence estimates rather than weight the results from the 1995 sample.” (Volberg 1995b, p. 5). We followed this precedent in past studies and do so again here. In the 1997 report, we produced both weighted and un-weighted estimates. Weighting did affect estimates of gambling problems in Michigan, though the magnitudes tended to be of a half a percentage point or less. As explained below, it is important to remember that response bias, to the extent that it is present in all gambling prevalence surveys of this type, almost certainly works to produce underestimation relative to the actual rates of gambling and problem gambling in the population.

Table 1 shows the characteristics of respondents to the 2006 Michigan survey and of Census descriptors for Michigan’s adult population. As expected, the statewide sample under-represents males, minorities, and the youngest, least educated, and poorest residents of the state. This selection bias is largest for gender though using a weight for gender had only a modest effect on the statewide prevalence rate (about a tenth of a percentage point). As before, weighted estimates are not reported because of their small effect and the lack of such practice in other studies. The final reason for using un-weighted estimates is that the assumptions of weighting (principally that non-respondents of a particular demographic category are well represented by respondents of that category) are rarely justified.

Table 1. Percent of the Sample in Demographic Categories Compared to Those of the 2000 Census Population Aged 18+

| |Statewide |2000 Census |

| |Sample | |

| N |957 |7,342,677 |

|Gender | | |

| Male |42.6 |49.0 |

| Female |57.4 |51.0 |

|Race/Ethnicity | | |

| White/Caucasian |85.9 |80.2 |

| Black/African |10.7 |14.2 |

|American | | |

| Other |3.3 |5.6 |

| | | |

|Hispanic |2.1 |3.3 |

|Age | | |

| 18-20 | 1.6 | 5.8 |

| 21-64 |71.3 |77.6 |

| 65 or older |27.1 |16.6 |

|Education | | |

| ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download