7. Discussion

The 2001 National Household Survey on Drug Abuse (NHSDA) represents the first year for which the sample sizes have been sufficiently large to estimate change in substance use among States based on the difference of two moving averages. Estimates from the 2000 NHSDA State small area estimation (SAE) report (Wright, 2002a, 2002b) established the utility of combining 2 years' data in order to better reflect the true variation among States. Estimates of change for this 2001 SAE report are based on modeling the difference between the averages for 2000–2001 and 1999–2000.

In addition to the new capability of measuring change, the survey provided estimates of serious mental illness (SMI) for the first time in 2001. That information provides the first picture of the distribution of SMI among the States.

7.1. Change in State Estimates

Change in substance use among States between 1999–2000 and 2000–2001 was estimated for 12 measures ranging from ones with low prevalence rates, such as cocaine use in the past year, to those with relatively high rates, such as past month use of alcohol. Because the size of the change is typically small, the focus has been on tests that indicate whether the change is significantly different from zero. The estimates of change considered in this report are differences in the prevalence estimates for 1999–2000 and 2000–2001. Results are presented when these differences are significant at the 0.05 or 0.10 significance level. However, the prediction intervals (PIs) are generally wide, and one should only characterize the change as being positive or negative. Regarding these significance levels, it is noted in Section 7.4 that the p values presented in the Appendix A tables are somewhat conservative (i.e., too large) because the associated correlation between the two overlapping prevalences was underestimated. A more precise significance level calculation presented in Appendix E yielded p values that were reduced by a multiple that ranged from 0.94 to 0.79 depending on the substance measure.

The estimated change between 1999–2000 and 2000–2001 is probably best interpreted as the average yearly change between 1999 and 2001 (see Appendix E for more details). The size of the change that was statistically detectable depended on a number of factors, including the level of the State prevalence rate, the size of the State sample, the fit of the national model, and the magnitude of the change itself. For past year cocaine use among persons age 12 or older, where the national prevalence rate was only about 1.5 percent in 2001 (Office of Applied Studies [OAS], 2002d), none of the State differences was statistically detectable, in part because the largest change was only 0.6 percent. In the same age group, for past month use of an illicit drug (6.3 percent in 2000), changes were detectable that were approximately 1.5 percent or larger for the States with annual samples of 900, and about 1.0 percent or larger for the States with large annual samples of about 3,600. For past month use of alcohol, a measure for which the national average among persons age 12 or older was close to 50.0 percent, changes of about 4.0 to 5.0 percent or larger were detectable. This is due in part to the fact that the variance of a percentage reaches its maximum at 50 percent.

The estimates for substance use among States for 2000–2001 were similar to those for 1999–2000. In general, the correlation between the two sets of estimates was quite high, about 0.80 or higher for each of the component age groups, and for all persons age 12 or older, except for past year use of cocaine and past year use of any illicit drug other than marijuana (26 or older). State rankings remained fairly constant over the period even though a number of States experienced statistically significant changes.

Most of the State-level changes for a specific measure mirrored the corresponding direction of the national trend for that measure. Occasionally, there were individual States that "bucked" the national trend.

Six States had significant increases of current use of illicit drugs among those age 12 or older. Three of those States ranked in the top fifth (i.e., had the highest rates) of illicit drug use (Maine, New Hampshire, and Vermont), two States were in the second highest fifth, and the other State fell into the middle fifth. Similarly, eight States recorded significant increases in current marijuana use among persons age 12 or older. Five of those States were among the top two fifths in current marijuana use. Delaware, a top fifth State for past month marijuana use among youths age 12 to 17 in 1999–2000, had the only significant decrease in that age group (also the only significant decrease in any of the age groups). Both Illinois and Vermont had significant increases between 1999–2000 and 2000–2001 in the incidence of marijuana use among persons age 12 or older, from 1.6 to 1.8 percent for Illinois and from 2.3 to 2.7 percent for Vermont, foreshadowing possible future increases in the prevalence rates for marijuana (Table A.4).

Although a number of States showed significant decreases in perceived risks of monthly use of marijuana (consistent with the increase at the national level), among those age 12 or older, only Alaska had a significant increase in perceived risk. The relationship between perceived risks of a substance and its actual use are negatively correlated at the person level; however, at the State level, changes in actual use were not as sensitive to changes in perceptions of risk. For example, Alaska's perceived risk of monthly marijuana use among persons age 12 or older increased during the period, and its actual use in that age group was slightly higher as well, though not statistically significant. Similarly, five States had significant decreases in perceived risk of monthly marijuana use, but only New York reported a significant increase in past month marijuana use in the same age group.

Given the small prevalence rate for past year use of cocaine and the size of the sample, it is no surprise that there were no significant changes among persons age 12 or older. Across all of the age groups, only Hawaii showed a significant change—a decline from 1.5 to 0.8 percent in the 26 or older age group. Hawaii had ranked in the highest fifth in 1999–2000 among persons age 26 or older, but fell to the lowest fifth in this age group for 2000–2001 (Table A.6).

Consistent with the national increase in current use of alcohol, nine States had significant increases between 1999–2000 and 2000–2001 among persons age 12 or older. Four of those States (Vermont, Connecticut, Minnesota, and Rhode Island) were ranked in the top fifth both in 1999–2000 and 2000–2001. Only Alaska, which ranked in the next to highest fifth in 1999–2000, showed a decrease among persons age 12 or older (from 52.9 to 49.4 percent) (Table A.7).

Recent use of tobacco and cigarettes at the national level among persons age 12 or older was reasonably flat between 1999 and 2001. Only four States reported any significant changes between 1999–2000 and 2000–2001 in past month use of cigarettes. Three States reported decreases in the 12 to 17 age group: Arkansas (from 18.6 to 14.6 percent), Oregon (from 15.0 to 11.8 percent), and Pennsylvania (from 16.4 to 14.8 percent). Rhode Island reported the only increase in current use of cigarettes—from 37.3 to 42.7 percent among persons age 18 to 25. Arkansas had ranked in the top fifth of States for youths age 12 to 17 in 1999–2000, but ranked in the middle fifth in 2000–2001 (Table A.11).

Perceptions of the risk of heavy smoking at the national level increased between 1999 and 2001 among persons age 12 or older and in each of the component age groups. Reflecting this trend, 15 States also had significant increases in perceived risk among persons age 12 or older during that period. Interestingly, none of these States indicated any significant decreases in current use of cigarettes, nor did any of the States that had significant changes in current use of cigarettes show correspondingly significant changes in perceptions of the risk of heavy smoking. Year-to-year changes in perceptions of risk of using a substance at the State level are not necessarily associated with corresponding changes in the actual use of a substance.

7.2. Characteristics of Prevalence Levels among States

State estimates of the prevalence of substance use can provide, among other things, information on the geographic clustering of these problems. Many factors can influence the nature of State and local prevalence rates, including local culture and social norms, State and local policies, and the sources, supply, and marketing of drugs. The findings in this report reveal varying degrees of clustering of substance use depending on the substance.

States with the highest prevalence of illicit drug use include five Northeastern States, four Western States, and one Southern State (Figure 2.1; Table B.1). By contrast, there was greater State clustering associated with alcohol and tobacco use. The highest rates of both binge alcohol use and general alcohol use were found in Northern States. The highest rates of past month cigarette and tobacco use were in the South.

Substance use literature has documented the inverse relationship between the perceptions of risk in using a substance and the actual use of the substance at the individual level (e.g., Bachman et al., 1998). The lower the perception that use involves risk, the higher the probability of use. This relationship at the individual level is reflected to varying degrees in correlations at the State level. Binge alcohol use provides an example of a "weak" relationship at the State level. Five out of ten States with the lowest percentages of perceived risk of binge drinking reported the highest levels of binge alcohol use (Figures 3.5 and 3.9; Table s B.8 and B.9). A similar relationship occurred between past month binge use of alcohol and past month use of alcohol in general, with five of the States that ranked highest in past month binge alcohol use also ranking highest in past month alcohol use (Figures 3.1 and 3.5; Table s B.7 and B.8).

A slightly stronger relationship with respect to the perception of risk and prevalence of use was found with cigarettes. Six States that had high rates of cigarette use also had the lowest rates of perceived risk of heavy use of cigarettes (Figures 4.5 and 4.9; Table s B.11 and B.12). The strongest relationship was found between perceived risk of occasional use of marijuana and past month use of marijuana. Nine of the States with the lowest perceived risk of occasional marijuana use also had the highest rates of past month use of marijuana (Figures 2.5 and 2.9; Table s B.2 and B.3).

Because marijuana is the most commonly used illicit drug, most of the States with the highest rates of illicit drug use also were the States with the highest rates of past month marijuana use (Figures 2.1 and 2.5; Table s B.1 and B.2). States where the rate of first-time use of marijuana was high also tended to be States with the highest rates of past month marijuana use although the correlation was somewhat less than one might expect (Figures 2.5 and 2.13; Table s B.2 and B.4). Of the 10 States in the top fifth with respect to past month use of an illicit drug, 6 were in the top fifth for past month use of an illicit drug other than marijuana (Figures 2.1 and 2.16; Table s B.1 and B.5). Seven of the States with the highest levels of past month use of illicit drugs other than marijuana also had the highest rates of past year use of cocaine (Figures 2.16 and 2.20; Table s B.5 and B.6). In general, a State that had a high level of use of one substance also tended to have high levels of use of related substances.

States that ranked high for substance use by all persons age 12 years or older also ranked high in use of substances by the population age 26 or older. This relationship derives from the fact that the latter group represents 77 percent of the total population 12 years old or older. Although the 26 or older population often drove the prevalence rates in the 12 or older population in a State, rates among the 12 to 17 and 18 to 25 age groups may not have followed suit. For example, California displayed a high rate for past month illicit drug use among all persons age 12 or older, but the rates in the 12 to 17 and 18 to 25 age groups were similar to the national average (Figures 2.1 to 2.3; Table B.1). On the other hand, Massachusetts, Vermont, Colorado, and Rhode Island had high rates of use of any illicit drug among all three age groups.

Another possible inference that can be made by comparing the States that displayed the highest rates of substance use among youths age 12 to 17 from 1 year to the next with the States having high rates in the 26 or older age group is that the behavior of the former group is more susceptible to change. The younger age groups represent ages of initiation and experimentation and groups that are probably more susceptible to influence and change; older persons' drug behavior is more established, with most former substance users having stopped.

With 2 years' data using the same definitions of dependence and abuse for six prevalence measures, the range of prevalence rates between the State with the lowest rate and the State with the highest rate is generally larger than it was with only a single year's data. This is in part a result of having sample sizes approximately twice as large so that the States' sample data carry more weight relative to the national model than they did with a single year's data. In 2000, the weight of the national model tended to pull down the States with high sample-based estimates. In 2000–2001, the sample-based estimates are relatively more precise and have been given more weight in the composite estimator. Except for dependence or abuse of any illicit drug, for which the range remained the same, the range of the other measures increased anywhere from approximately 20 to 50 percent, usually at the top of the range. The higher ranges also are the result of increased prevalence rates between 2000 and 2001 for some of the measures.

From 2000 to 2001, the national percentage of those with dependence or abuse increased for both illicit drugs and alcohol (OAS, 2001b, 2002c). The relationship between past month use of alcohol or past month binge use of alcohol to past year alcohol dependence or abuse was not particularly strong due in part to the widely different prevalence levels of the measures. For example, among the States with the highest rates of current alcohol use for those age 12 or older (States ranged from about 55.1 to 61.6 percent), only four States fell into the highest fifth for past year dependence on or abuse of alcohol (rates ranged from 6.8 to 8.5 percent). Even with respect to the smaller percentage of past month binge use of alcohol, only 5 States in the top 10 for binge alcohol use also were present in the top fifth for alcohol dependence or abuse in the past year (Figures 3.1, 3.5, and 5.1; Table s B.7, B.8, and B.13).

Only four States ranked in the top fifth for past year alcohol dependence or abuse also were ranked in the top fifth for past year alcohol dependence: the District of Columbia, New Mexico, South Dakota, and Alaska. For the States in the top fifth for past year alcohol dependence, the percentage of persons age 12 or older who met the criteria for dependence comprised anywhere from 38 to 49 percent of those meeting the criteria for both past year dependence or abuse. Although the top States for current use or binge alcohol use were primarily States from the northern parts of the United States, the top States for past year alcohol dependence included more Southern, Western, and Midwestern States: the District of Columbia, Louisiana, Mississippi, Oklahoma, Illinois, South Dakota, New Mexico, California, Oregon, and Alaska (Figures 3.1, 3.5, 5.1, and 5.5; Table s B.7, B.8, B.13, and B.14).

Generally, States with high prevalence rates for alcohol dependence or abuse were not the same States that had high prevalence rates for illicit drug dependence or abuse. Only three of the States in the top fifth with the highest rates of alcohol dependence or abuse (Massachusetts, New Mexico, and Colorado) also were in the group of States with the highest levels of illicit drug dependence or abuse (Figures 5.1 and 5.9; Table s B.13 and B.15). Most of the States with the highest levels of illicit drug dependence or abuse were in the West: Nevada, California, Washington, New Mexico, Colorado, and Oregon. The top fifth also included one State from the South, Louisiana, and three from the Northeast: Massachusetts, Connecticut, and Vermont. Only two States were in the top fifth for all three age groups: Nevada and Connecticut.

There was some degree of relationship between high rates of past year illicit drug dependence or abuse and high rates of past year cocaine use at the State level. Six States were ranked among the highest for both measures: Colorado, New Mexico, Massachusetts, Nevada, Vermont, and California (Figures 2.20 and 5.9; Table s B.6 and B.15).

Not only did geographic clustering of States occur among those with high prevalence rates, but similar clustering also was evident among the States with the lowest rates. For example, nine Southern States were in the lowest fifth for past month use of alcohol (Figure 3.1; Table B.7), eight Southern States were in the lowest fifth for past month binge use of alcohol (Figure 3.5; ), and seven Southern States were among those indicating a high risk of binge drinking (population age 12 years or older) (Figure 3.9; Table B.9). By contrast, only one Southern State was in the top fifth for current use of alcohol, and no Southern State appeared in the set of States with either the highest rates of binge alcohol use or the lowest rates for perceived risk of binge drinking. Similarly, 10 Southern States comprised the category of States with the highest perceived risk of using marijuana occasionally, but only Delaware was in the group of States with the lowest perceived risk of marijuana (Figure 2.9; Table B.3). Also, six Midwestern States were among those indicating the lowest rates of past year dependence on any illicit drug; however, no Midwestern State was among those with the highest rates of illicit drug dependence (Figure 5.13; Table B.16).

The estimates of the percentage treatment gap for 2000–2001 displayed a larger range of percentages, especially among States with the highest percentages. Part of the reason for this was the increase in the percentage treatment gap at the national level from 1.7 percent in 2000 to 2.2 percent in 2001 (OAS, 2002c). The other part is due to the nature of the estimation process that gives relatively more weight to the sample data for 2000–2001 relative to 2000 because there are 2 years of data instead of 1. The precision of the estimates as indicated by the smaller PIs also has improved. California had the largest percentage treatment gap in 2000–2001, and the other States in the top fifth were mostly from the West or Northeast (Figure 5.21; Table B.18). Along with the national increase in the percentage treatment gap between 2000 and 2001, it can be inferred that most States shared in that increase. States in the two lowest fifths in 2000 had increases that were on average lower than the national average, States in the middle fifth had increases that were similar to the national average increase, and most States in the highest two fifths displayed increases that were greater than the national average.

7.3. Serious Mental Illness

The 2001 NHSDA was the first in which the survey was capable of providing estimates of SMI for all persons age 18 or older. States with the lowest rates of SMI were a mixture of one Western State, three from the Northeast, three from the South, and three from the Midwest (Figure 6.1; Table B.20). The State with the lowest rate was Hawaii (5.1 percent). States in the highest fifth seemed more clustered geographically with six Southern States, three Western States, and one State from the Midwest. Oklahoma, the State with the highest rate of SMI, had a rate that was double that of Hawaii. Estimates of SMI among the States with larger samples fell into a narrower range: from Florida at 6.8 percent to Michigan with 8.2 percent. Persons age 18 to 25 had higher rates of SMI than did the 26 or older age group. In the 18 to 25 age group, California had the lowest rate (9.7 percent) and Maine had the highest rate (14.4 percent).

Although SMI is somewhat correlated at the individual level with past month use of an illicit drug, the correlation at the State level was fairly low and negative (-0.18). The highest correlation at the State level was between SMI and past month use of cigarettes, 0.31. This result is supported somewhat by substance use literature that shows a relationship between SMI and past month use of cigarettes at the individual level (Arday et al., 1995; Kessler et al., 2003; Romans et al., 1993; Woolf et al., 1999). The correlations with dependence on or abuse of drugs or the need for treatment were generally quite low. The highest correlation with demographic information was with the 1999 per capita income obtained from the Bureau of Health Professions' 2002 Area Resource File, where the correlation was -0.53: the lower the income, the higher the percentage with SMI.

In general, the State estimates derived from the NHSDA data correlated only moderately, 0.259, with the synthetic State estimates generated from the Epidemiologic Catchment Area (ECA) study and the National Comorbidity Study (NCS) and published in the Federal Register by the Center for Mental Health Services (CMHS, 1999). The data used from the ECA were limited to Baltimore and were collected during the 1980s. The NCS data were from a national probability sample of approximately 8,000 households and included data for only 34 States. The method used was essentially based on synthetic estimation in which the NCS data were used to make estimates for persons 15 to 54 years old, and the ECA data were used to make estimates for persons age 55 or older. The estimation used a fixed-effect logistic regression model based on data at either the county or Census tract level consisting of demographic information, such as age, race/ethnicity, and gender. By contrast, the State-level SMI estimates in this report are based on representative State samples of about 2,400 persons for the eight largest States and 600 persons for the 42 smaller States and the District of Columbia surveyed throughout the 2001 calendar year. The NHSDA model includes random effects at the State and field interview region group levels in order to reflect differences among States and region groups that are not captured by the fixed-effect national model.

7.4. Validation

It is difficult to find other data to validate the State-level estimates discussed in this report and presented in the tables. In the past, national estimates from the NHSDA have been compared with estimates from the Behavioral Risk Factor Surveillance System (BRFSS) and the Youth Risk Behavior Survey (YRBS) sponsored by the Centers for Disease Control and Prevention (CDC, 2003a, 2003b). However, these CDC surveys (a) did not focus extensively on substance use, (b) employed different data collection methods, (c) did not cover all of the States on an annual basis, and (d) had varying degrees in potential response and nonresponse bias. It is, therefore, difficult to know how much confidence should be placed in comparisons of results.

Although external validation of NHSDA findings is problematic, internal validation of the States can be useful. Because the State prevalence levels for 2000–2001 are estimated in the same manner as they were for 1999–2000, the procedure for, and results of, that validation are first summarized here from last year's report (for details, see Volume II, Section B.4.2 in Appendix B, in the 2000 State report [Wright, 2002b]). Subsequently, the process for validating the estimates of change between 1999–2000 and 2000–2001 are presented, as well as the results of that analysis.

To validate the modeling process for estimating the State prevalence levels for 1999–2000, data from 1999 and 2000 were combined for each of the eight largest States, resulting in sample sizes of about 7,200 per State. Given the large sample sizes and the precision of estimates based on samples of this size, the sample estimates for each of the eight States were considered to be the true values. Replicating the sample design and model estimation procedures used in producing small area estimates for the 42 States and the District of Columbia (based on pooled samples of about 1,800 persons age 12 or older), each of the eight large States was divided into four "pseudo" substates. Estimates then were produced for four substance measures and three age groups for each of the four "pseudo" States (see Table s E.13 to E.16 in Appendix E in Volume II). Comparing the results with the true values in each of the eight States, the State model estimates (for all persons age 12 or older) were very close to the true values (i.e., the bias as a percentage of the estimated prevalence rate was very small):

past month use of marijuana, 4.07 percent;
past year use of cocaine, 7.88 percent;
past month binge alcohol use, 0.98 percent; and
past month use of cigarettes, 1.22 percent.

For example, if the true value of past month use of marijuana for persons age 12 or older in a State with a pooled sample of about 1,800 persons was 5 percent, the small area estimate would, on average, fall within 0.2 percent (4.07 percent × 5 percent) of the true value. In addition, the range within which the true value will lie 95 percent of the time (referred to as the 95 percent prediction interval [PI]) was much smaller than the corresponding estimate based solely on the sample: 25 to 35 percent shorter, depending on the substance (Table E.17). Taking both the model bias and the range of estimates into account, the total relative standard error was quite small-much smaller than could be obtained using just the sample.

The validation process for measuring change was similar to that for measuring prevalence levels and is described in more detail in Volume II, Appendix E, of this report. The change measure is defined as the odds ratio {P2/ (1-P2)}/{P1/ (1-P1)}, where P1 is the pooled 1999–2000 small area estimate and P2 is the pooled 2000–2001 small area estimate.

The results of the validation study indicate that the relative absolute bias of change between 1999–2000 and 2000–2001 was fairly small (see Table s E.9 to E.12 in Appendix E, Volume II), but somewhat larger than it was for the estimates of the 2000–2001 prevalence levels. For the population age 12 or older, the average relative absolute bias of change is as follows:

past month use of marijuana, 4.26 percent;
past year use of cocaine, 12.71 percent;
past month use of alcohol, 3.28 percent; and
past month use of cigarettes, 1.42 percent.

Generally speaking, the model-based estimates tend toward a conservative underestimate of change. That is, the true change for the above four substance measures, whether an increase or a decrease, tends to be larger than the model-based estimate. Because the validation is focused on the States that have sample sizes of approximately 1,800 persons, the expectation is that estimates for the eight large States, where the sample sizes are approximately 7,200 (for pooled data across 2 years), would be closer to their true values than is indicated for the four substance measures presented above.

The p values presented in Appendix A's tables are somewhat conservative because the model-based approximation of the correlation between 1999–2000 and 2000–2001 that was used in those tables underestimates this quantity. In a separate analysis, the size of the underestimate has been estimated for the above four substance measures (see Table s E.5 to E.8) using an alternative estimator that is more precise. (For more details on the reestimation of the correlation, see Volume II, Appendix E.) The underestimate appears to be more prominent for the measures with higher prevalence rates, including past month use of marijuana, past month use of alcohol, and past month use of cigarettes. The ratios of the average reestimated p values to the average original p values for the 12 or older population are 0.79 for past month use of alcohol, 0.81 for past month use of cigarettes, 0.84 for past month use of marijuana, and 0.94 for past year use of cocaine. Table E.3 presents the ratios of the average width of the model-based PIs based on the reestimated year-to-year correlations relative to the average width of the design-based confidence intervals of change for the substate areas for the four validation substance measures. Based on the reestimated correlations, the true ratios of the model-based interval widths to the design-based interval widths range from 0.60 to 0.77 for the 12 or older population, somewhat lower for the low prevalence measures, cocaine and marijuana, than for alcohol and cigarettes. These ratios imply that, for small States, the model-based estimates result in gains in precision over the usual direct sample-based estimates that are equivalent to sample sizes that are approximately 1.7 (1/.77²) to 2.8 (1/.60²) times as large as the actual sample sizes, depending on the substance.

However, the model may not be able to adequately adjust for differential nonresponse and bias effects at the State level. There were considerable differences in the response rates between States with the lowest and highest rates. In 1999, for example, Massachusetts had the lowest response rate at 49.8 percent and Mississippi had the highest rate at 78.2 percent. In 2000, the range of response rates was somewhat smaller with the Illinois rate at 58.2 percent and the Kentucky rate at 80.6 percent. In 2001, the overall response rates at the State level ranged between 55.3 percent for Illinois and 78.5 percent for New Mexico. If there were bias resulting from nonresponse that varied in relation to the rates, it would raise questions about comparisons among States. (See Volume II, Table s E.18 to E.20 in Appendix E, for interview response rates by State in 1999, 2000, and 2001.)

There was, in fact, some suggestion that the State nonresponse rates and the prevalence levels of substance use were somehow related. Averaging State response rates for the 1999 NHSDA and the 2000 NHSDA and comparing the result with the rate of past month marijuana use by persons 12 years or older (using the pooled 1999–2000 data) revealed a -0.42 correlation, suggesting that lower State response rates may be associated with higher State marijuana prevalence rates. This result is not sufficient to conclude there was in fact nonresponse bias. For such bias to exist after nonresponse adjustments have been made requires that the true probabilities for persons to respond to the survey still depend to some degree on whether they have used a substance or not.

Research has shown that the more socially unacceptable the substance, the greater the tendency to not report its use (Harrison, 1997). Therefore, one might anticipate very little underreporting if the question asked whether the respondent had ever used marijuana during his or her lifetime, but more extensive underreporting if asked about past month use of heroin. Some of the uncertainty about the extent and nature of the underreporting is being addressed by a validity study using hair and urine samples provided by respondents in the NHSDA.