> W@ bjbjFF +,,2.4$xxxsss8stl`xy"zzz%{%{%{'))))))$ТR"`MQx!{%{MXXzzێێێX@zxz'ێ'ێێxێzTxbqs'zێ+
0ێێXXXXxێP%{;~4ێo3%{%{%{MMQPXD"PX
Replicate Measures: A latent class analysis of measurement error
Submitted by:
J. Carl Setzer
University of Maryland, JPSM
Spring, 2004Introduction
In an effort to assess the accuracy of survey estimates, researchers often compare estimates obtained from two or more sources. For example, researchers will sometimes compare administrative records with estimates obtained from a survey (Groves, 1989). In such cases, the administrative records are usually assumed to be more accurate, or what are often referred to as the gold standard. Other times, researchers compare estimates against benchmarks, such as census or CPS data, which are assumed to be more accurate descriptions of the population due to better coverage. Others still use reinterviewing techniques, where a respondent is asked the same question or set of questions on separate occasions. When differences between the first and second measures are found, the respondent is recontacted and reinterviewed in order to reconcile the inconsistent responses. These methods can be quite costly, impractical, or exceedingly burdensome not only to the researcher, but also to the respondent. One other possible method, called repeated (or replicate) measurement suggests replicating similar questions, not over repeated trials, rather within the same survey or interview (Groves, 1989). This method is somewhat similar to reinterviewing the respondent or the testretest method. This technique is often used in psychometrics and educational testing for the purpose of measuring a persons attitudes, behaviors, or knowledge. Slight changes in the wording of the questions are needed in order to mask the intentions of the researcher.
In theory, repeated measures offer the analyst an opportunity to look closely at measurement error among the items. When a set of questions are used to measure one particular trait, they are often referred to as parallel measures if it is expected that the measurement error (or precision) is the same for each of the items. Following the notation used by Groves (1989), parallel measures can be described using the following equation:
EMBED Equation.3 and EMBED Equation.3
where yikm and yikm are indicators (m and m) of the kth underlying characteristic for the ith person; Xik is the true value of the kth characteristics for the ith person; and im, im are the random error terms for the two indicators. In order for the measures to be parallel, the following assumptions must apply: E(im)=E(im )=0, meaning that the indicators are measuring the same thing and have the same expected value; Var(im) = Var(im ), or the indicators have the same variance and measure the underlying characteristic with the same amount of precision; and finally Cov(Xij, im) = Cov(Xik, im ) = 0, the error terms and true values do not covary. The discussion by Groves also mentions that the error associated with each of the measures is not correlated, which is similar to the axiom of local independence (discussed further below). It is noted by Groves that this assumption is quite restrictive and may not be met. For example, one may expect that the correlation between the errors will be negligible when the items are placed further apart in the questionnaire. Items closer together, however, may tend to show stronger correlations. This may be particularly true if the items are subject to a social desirability bias. In the presence of social desirability, the respondent may provide responses that deviate from their true score, and given the indicators are replicate measures, their errors are more likely to be correlated. That is, if the respondent is concerned about telling the truth, this concern is likely to be reflected in all the items. In addition, the fact that the replicate measures are alternately worded may suggest that the error distributions are unequal, violating the assumption that the items are parallel (as compared to being equal if the exact same question was asked multiple times in the same instrument).
As mentioned earlier, replicate measures found within the same survey are often employed to improve the accuracy of an estimate. For example, the National Household Survey on Drug Abuse (NHSDA) used repeated measures to estimate marijuana usage as recently as 1999. The 1999 NHSDA asked a set of questions about recency, frequency, and other behaviors related to marijuana use. The overall prevalence rate was calculated using information from this set of questions, rather than relying on a single indicator. Specifically, respondents who answered positively to one of two questions were classified as users. This, however, introduces the possibility of false positive and false negative responses, i.e., measurement error. Biemer and Wiesen (2002) took a different approach to the NHSDA data, utilizing latent class analysis (LCA) to compensate for false positive and false negative responses and to produce a more accurate estimate. Furthermore, the authors tested the hypothesis that the indicators, following a Markov process, were locally independent; that is, they tested whether the indicators were correlated. Note here that the indicators used in the analysis were located in different parts of the survey, rather than together. Contrary to their hypothesis, the latent class model ultimately selected did not allow for any correlations between the indicators (beyond that captured by the latent variable), supporting the conclusion that the assumption of local independence was adequate. It was also found that the true prevalence rate of marijuana usage varied by age, race, and sex.
Biemer and Wiesen leave open (and untested) the possibility that items in the same section of a survey (i.e., consecutive items) will show a stronger correlation than items spaced further apart. Therefore, like Biemer and Wiesen, the present study attempts to model replicate measures using an LCA approach. It is hypothesized, however, that items will be correlated with the first question in the set. In other words, how a respondent answers the first item influences how he or she responds to subsequent items in order to keep their responses consistent. Their approach, which is typical of LCA modeling, was to begin with a basic model and introduce correlated error terms and higherorder interactions. The first model should test for zero correlations among the errors for any two indicators. For the present study, data from the Youth Risk Behavior Survey (YRBS) will be used to test the assumption of local independence between a set of indicators. Following a brief discussion on latent class analysis, we describe the YRBS survey and the data.
Latent Class Analysis
Latent class analysis is a technique that is very similar to factor analysis, the exception being that the latent variables are categorical in nature rather than following a continuous scale. It is in the class of general linear models, in that it is an extension of a loglinear model utilizing both latent and manifest variables. LCA is used to determine whether the cases can be classified into two or more different latent categories. The theoretical background and model development methods are well documented in the literature (see McCutcheon, 1987; Hagenaars, 1993). For illustration, the general form of a saturated probabilistic model with four indicators (A, B, C, D) and T categories of the latent characteristic, X is shown below,
EMBED Equation.3 , where EMBED Equation.3
EMBED Equation.3 denotes the probability that in the population a randomly selected individual scores (i,j,k,l) on the joint variable ABCD. The probability of obtaining the score (i,j,k,l,t) on the joint variable ABCDX is indicated by EMBED Equation.3 , and is the probability of belonging to class t of X. The parameter EMBED Equation.3 is a conditional response probability, namely, the probability that an individual obtains the score A = i, given this person belongs to latent class t of X. In order for the model to be identified, the parameters are restricted so that, for example, EMBED Equation.3 , which indicates that both the sums of the latent class probabilities and the conditional probabilities are one. The above model is shown in a probabilistic form. However, the model can also be written in terms of a loglinear model, such as:
EMBED Equation.3
or,
EMBED Equation.3 (1)
In shorthand, the above model can be written (in loglinear form) as {AX, BX, CX, DX}. This is the standard model that incorporates the assumption of local independence.
In addition to the above restrictions, the axiom of local independence further restricts the indicators to have zero partial associations with each other. In equation (1) above, this is reflected in the absence of interactions between any of the indicators. However, Hagenaars (1988) described situations when this assumption can be relaxed, allowing for the possibility of correlation between indicators. There are two types of this socalled local dependence model  symmetric and asymmetric dependence. These two variations are best described using figures 1 through 3. Figure 1 represents the axiom of local independence in the relationship between latent variable X and four indicators (A,B,C,D). The arrows represent the idea that each of the indicators is explained by its relationship to the latent characteristic. There are no direct effects between any two of the indicator variables. However, in Figure 2, we see an example of a symmetric model, where the local independence assumption is relaxed. Here, the double arrows between A and B represent the correlation between the two and here it is assumed that A and B are not causally related. Figure 3, on the other hand, shows an asymmetric relationship between A and B, where A is causally related to B. For example, if A is a question item in a survey, and B is a second item preceded by A, one could make a case that the response to A influences the response to item B. This may especially be true in the presence of social desirability, since one could expect respondents to try and be consistent in their responses, even if they are inaccurate.
EMBED FlowCharter7.Document EMBED FlowCharter7.Document EMBED FlowCharter7.Document
Figure 1. Figure 2. Figure 3.
The latent class model described above can be extended to test several samples, or grouping variables, at one time. For example, in the Biemer and Wiesen article, the authors tested their measurement model across age, race, and sex simultaneously. This allows the analyst to test for interactions among the indicators, latent characteristic, and grouping variables. The general probabilistic form of the model is,
EMBED Equation.COEE2
where gs is the conditional probability of an observation being in latent class g, given that it is in group s; and ixgs is the conditional probability that an observation from group s and class g will be at level x of indicator i. Restrictions to maintain identifiability are, EMBED Equation.3 (i.e., the conditional probabilities sum to 1) and EMBED Equation.COEE2 (i.e., for a given group, the probabilities of being in the different latent classes sum to 1). Here, X represents the latent class, while A, B, and C are three separate indicators, or manifest variables. The model can also be written using loglinear parameterization, with two levels of group s, and indicators Ah, Bj, Ck, and Dm as,
EMBED Equation.COEE2
where the following restrictions apply,
EMBED Equation.COEE2
where N is the superscript for the group effect. As before, the model can also be written in shorthand as {AXN, BXN, CXN, DXN}.
Youth Risk Behavior Survey
The Youth Risk Behavior Survey (YRBS) is sponsored by the Centers for Disease Control and Prevention and is designed to monitor behaviors that are related to causes of mortality, morbidity, and social problems among youth and adults (ages 12 to 21) in the United States. Individual states (and some local agencies) are encouraged to utilize the YRBS in order to identify problems that need to be addressed. Each state, however, is ultimately responsible for collecting its own data and sampling methodologies vary by state. In general, however, classrooms are selected from a sample of schools, and each student within the classroom is asked to participate (with parental consent).
The YRBS dataset used for the current analysis includes 2003 data from four states and one large city. The general process used to create the public use files is to perform edits and imputation, followed by weighting procedures. In the case of the YRBS dataset, however, inconsistent responses between specific sets of indicators, e.g. the questions regarding sexual activity are completely excluded from the weighting procedures and are eliminated from the final dataset. Because of this procedure, the dataset used for this current analysis is simply the raw data file, and excludes any weights.
The YRBS contains a set of seven items pertaining to sexual activity (see Appendix A for exact wording of items). The instrument has been designed to exclude any skip patterns. Because of this, respondents are provided with an opportunity to indicate he or she has never had sexual intercourse on each item, allowing them to be dichotomized into two categories: those who indicate they have never had sexual intercourse and those whose responses indicate that they have had sexual intercourse. In theory, one should expect these items to have consistent response patterns, although the original wording of the items differs and therefore the items may have different error distributions (i.e., the items are not necessarily parallel measures). In addition, because the items appear within the same section of the same survey and are consecutive, I expect that the items may have correlated errors, meaning that mistakes made on one item should be correlated with errors made on the other items. Also contributing to any correlation across items may be social desirability.
When the dataset was examined, it was discovered that a small number of respondents clearly did not make any effort to answer the questions correctly. For example, one respondent answered every item with the same response. Therefore, the dataset was reviewed prior to the analysis, and any obvious response patterns (e.g., same response across all survey items, some alphabetical pattern, etc.) were removed from the dataset. There were 14 records removed, resulting in a sample size of 5,834. The analyses are limited to the respondent population only.
Results
Gross Error Rates
Prior to performing the latent class analysis, comparisons of the gross error rates were conducted. The gross error rates (calculated as the sum of the offdiagonal cell frequencies, divided by the sample size for each pairwise comparison) for all twoway comparisons are provided in Table 1. The data show that in all but one case, adjacent indicators have lower error rates than nonadjacent indicators. For example, the error rate between Q58 and any other item is lowest when compared with Q59 (i.e., error rate = 1.37). The error rate increases from 1.37 percent for the Q58Q59 comparison to 1.73 percent for the Q58Q60 comparison, and so on. The greatest error rate occurs among the Q58Q62 pair, which is separated by three other items.
Two other trends can be seen in Table 1. One is that question 62 has the highest pairwise error rate with any of its prior items. Looking at the row for item 62 in Table 1, it can be seen that these error rates are higher, and all error rates decline subsequently. Second, when looking across the columns, the error rates for question 58 are higher than in any other column. These findings suggest that questions 58 and 62 may be more problematic. Potential reasons for this are provided in the next section.
Table 1. Gross error rates for twoway comparisons
Item5859606162636458591.37601.730.91611.801.071.09622.291.721.801.80631.981.351.411.481.16641.671.171.281.331.580.95Note: See Appendix A for exact wording of items.
After analyzing the gross error rates for the entire set of items, I examined the actual response patterns. Table 2 provides a summary of the discrepancies found among all possible response patterns, for all students and separated by gender. As expected, the vast majority of students provided consistent responses to the questions on sexual activity. Although not shown in the table summary, almost 97 percent of responses were consistent across all seven items. Of the patterns that were inconsistent, almost 82 percent were the result of a single discrepancy. Of these, 49 percent occurred with either question 58 or 62 (e.g., patterns such as NNNNYNN and YNNNNNN). Oddly, there were no occasions where a student gave discrepant responses to both questions 58 and 62. Overall, inconsistent responses at either question 58 or 62 account for 40 percent of the total number of errors. These results confirm that questions 58 and 62 may be problematic.
It is also interesting to note that males gave more than twice as many discrepant responses than females. Additionally, the response patterns for males included two discrepancies (22 percent) more often than female response patterns (10 percent). In other words, when males give inconsistent responses, they seem to do it across more than one item.
Table 2. Discrepancies in response patterns: frequencies and percentages
All RespondentsFemalesMalesDiscrepanciesFrequencyPercentFrequencyPercentFrequencyPercent One13481.74590.08777.7 Two +3018.3510.02522.3Total164100.050100.0112100.0Note: Row frequencies may not sum to total due to missing data.
Latent Class Analysis
I next attempted to model the measurement error of these replicate measures using latent class analysis. In particular, four of the seven indicators regarding sexual activity, along with a grouping variable for sex (i.e., gender) were tabulated in a crossclassification table. It should be noted that question 61 was omitted from the LCA models for several reasons. First, the reference period for the question is the previous three months, whereas the other items span the respondents lifetime. Second, and more importantly, I wanted to make sure that questions 58 and 62 were included in the analyses since they appear to be somewhat problematic. Including all five indicators would have resulted in a large number of sampling zeros, spreading the data too thinly for reliable estimates. When comparing the crossclassification table, item 61 resulted in many more sampling zeros than question 60 and was thus omitted.
Using sex of the respondent as a grouping variable allowed the ability to test whether the measurement error model is the same for both males and females. The grouping variable was included also because of the result that males tended to have higher numbers of discrepancies in their response patterns, as noted above. Previous research has studied and shown social desirability can produce different biases in sexual activity reporting for males and females (Tourangeau & Smith, 1996). Men tend to overreport, for example, the number of their sexual partners whereas women tend to underreport theirs. Using the grouping variable, though, decreased the size of the joint frequencies in the fully crossed table. Because of this, the same set of models was tested with and without the grouping variable for sex. Furthermore, models both including and excluding the correlated error terms were compared. Finally, the LCA model also allows the ability to estimate the classification errors (false positives and false negatives) and possibly identify weaker indicators of sexual activity. The LCA models were tested using LEM software (Vermunt, 1997).
To test a latent class model, several pieces of information are considered. First, the model must be identified, which means that there are more degrees of freedom than estimated parameters. Second, one seeks a model with a likelihood ratio goodnessoffit statistic (L2 ) value that is not significant i.e., the pvalue is greater than .05 (or .01 depending on factors such as sample size). Finally, in comparing nonnested models, one should look for the lowest BIC and AIC values. When the sample size is large, as is the case here, the BIC is particularly useful because it penalizes overparameterization by means of a logarithmic function of the sample size (Lin & Dayton, 1997). These three criteria were used to evaluate two sets of models. The first set tests for a single latent variable (restricted to two classes) while the second set introduces a grouping variable for respondent gender. The same approach was taken in both sets i.e., starting with a basic latent class model, which assumes local independence, followed by models that relax this assumption.
One final point to be made has to do with the cell counts. Because we could expect (and actually found) the number of gross errors to be small and also because the data are not weighted, several of the joint cell counts are small. In fact, when adding the grouping variable to the second set of models, several sampling zeros occur in the crossclassification table. To account for this, a common remedy is to add a value of 0.5 to each cell count (Knoke & Burke, 1980). The implications of the small expected frequencies are discussed further below.
Single latent variable model. The simplest model tested is the unrestricted latent class model, which assumes that the indicators are locally independent and that each is directly related to the latent variable, X (see figure 1 above). These are the typical assumptions made in an LCA and are often a starting point for the modeling procedure (Hagenaars, 1990). Here, it is assumed that the true score represents whether the respondent has ever had sex, which in turn directly determines how the respondent answers each of the four questions. In the absence of any measurement error we would expect this model to fit. However, from Table 3 it can be seen that model 1 does not fit the data well, with an L2 value of 108.6, df=6, and pvalue of 0.0000.
The assumption of local independence is quite restrictive; I next examined models that relax this assumption and test whether the indicators have correlated errors. Model 2 tests the hypothesis that the manner in which a student responds to the first question on sexual activity influences how he or she responds to subsequent items. This corresponds to the AB, AC, and AD parameters. In addition, model 2 allows for a direct relationship between BC, and CD. In other words, model 2 tests the hypothesis that each item is correlated not only with the first one (i.e., question 58) but also with the item immediately preceding it. Model 2 fits the data well. The L2 value is 0.1716, and the pvalue is equal to 0.6787, although only one degree of freedom remains.
EMBED FlowCharter7.Document EMBED FlowCharter7.Document
Model 2 Model 3
Single latent variable model with grouping (simultaneous LCA model). As mentioned, models 1 and 2 only test within a single sample. To test the hypothesis that males and females respond differently to the same set of questions, a grouping variable for gender was added to each of the remaining models. Model 3 in Table 3 is therefore similar to model 1, with the addition of the gender term or grouping variable, S. This is what is referred to as the group homogeneity test, because here it is assumed that although the latent class probabilities may vary by gender, the indicators do not. Model 3, which has L2 = 163.11, df=20 and p = 0.0000 does not fit the data well and indicates that the error distributions do vary by sex for at least one of the indicators.
EMBED FlowCharter7.Document EMBED FlowCharter7.Document
Model 4 Model 5
Next, threeparameter interactions were tested among each of the indicators, the latent variables, and the grouping variable. Model 4 allows the error distributions for each indicator to vary simultaneously among gender and the latent characteristic, in addition to the AB,AC,AD, BC, and CD parameters. However, model 4 resulted in a boundary issue, which is likely the result of extremely small cell sizes (or even empty cells). When a socalled boundary estimate occurs, it is likely because the expected cell frequencies are too low to render any reliable parameter estimates. Given the higherorder interaction and additional parameters, this is a plausible assumption. Because of the boundary issue, the resulting parameter estimates associated with model 4 are unreliable.
Models 1 and 3 did not fit the data well, and model 4 resulted in boundary estimates. Model 2, however, fit well and is consistent with earlier findings. Although this model appears to be an adequate representation of the relationships among indicators, some caution is warranted. First, the model is complex, resulting in a single degree of freedom. Second, according to previous research (see for example, Poon, Tang, & Wang, 2003; Reiser & Lin, 1999), when cell counts are sparse the chisquare approximation for traditional goodnessoffit statistics (e.g., the likelihood ratio) will not be valid. In addition, the parameter estimates and ultimately the conditional probabilities may be affected. Although the data technically are not sparse, the likelihood ratio and conditional probabilities may still be slightly affected by small expected cell frequencies.
Two latent class variables. Although model 2 appears to be the best model, one final hypothesis was tested. Instead of testing for direct relationships between each of the indicator variables, one could imagine that a second latent variable may contribute to the association between the indicators. It was hypothesized, then, that along with the true score latent characteristic, X, a second latent variable for social desirability, Y, would also account for any relationships between the indicators. Model 5 in Table 3 represents this hypothesis. However, when testing the hypothesis that social desirability has only two levels, at least one model parameter resulted in a zero estimate. Therefore, no further models with a second latent variable were tested.
Table 3. LCA Models
Modeling 58596062DFNumber of parametersL2pvalueBICAICModel 1 {AX, BX, CX, DX}610108.60.00005696Model 2 {AX, BX, CX, DX, AB, AC, AD, BC, CD}1150.17160.678781Modeling 58596062 by sexModel 3 {XS, AX, BX, CX, DX} group homogeneity test2012163.110.00009123Model 4 {AXS, BXS, CXS, DXS, AB, AC, AD, BC, CD}172512.480.0857471Modeling 58596062 and two latent variables, X and YModel 5 {AX, BX, CX, DX, AY, BY, CY, DY}11154.540.0329421 One parameter is nearly boundary and therefore unreliable.
Latent Class Output. Under the assumption that model 2 is correct, Table 4 shows the false positive and false negative error rates for each of the four indicators, given a true score value (as predicted by the LCM). From this it appears there is some overreporting of sexual behavior in the YRBS. The false positive rates are much larger than the false negative rates. The false positive rate for question 62 is greater than 4.5 percent, whereas the false negative rate is less than half of one percent.
Table 4. Latent Class Output: False positive and false negative rates
True ScoreObserved ScoreFalse Positives (percent)No (X=1)Q58=Yes (A=2)3.42Q59=Yes (B=2)2.99Q60=Yes (C=2)3.60Q62=Yes (D=2)4.53False Negatives (percent)Yes (X=2)Q58=No (A=1)0.47Q59=No (B=1)0.34Q60=No (C=1)1.41Q62=No (D=1)0.47
Discussion
The analyses contained in this paper further the research of Biemer and Wiesen (2002), who found that replicate measures located in separate sections of a survey did not necessarily have correlated error terms. The analyses above looked at replicate measures located in the same section of a single survey. Potential problems with two of the indicators were also found.
Because the items on sexual activity are located in a single section in the YRBS survey, I hypothesized that the error distributions of the items would be correlated. Two sets of models were tested where the difference was the inclusion/exclusion of the grouping variable for the respondents sex. After rejecting the two models that represented no correlations between the errors in the indicators, other models introduced parameters to test for local dependence. The final model, which supports the hypothesis of correlated errors, suggests that there is an anchoring effect with the first item, and also adjacent items. The respondents answer to the first question (i.e., question 58) affects responses to subsequent items on sexual activity. In addition, adjacent items appear to have an affect on subsequent indicators. This pattern is supported by previous literature that suggests how people bring their responses in line with one another for consistency. According to Converse and Presser (1986), for example, as a general ruleitems affect one another mainly when their content is clearly relatedor when the answer to one question has an obvious implication for the answer to another, (pg. 40). A followup question in this case is why people answer falsely in the first place. Although the answer to that question is beyond the scope of this analysis, a likely culprit is the social desirability effect (see Fowler, 1995; Tourangeau, Rips, & Rasinski, 2000). Students who feel to respond a certain way due to social pressures may be likely to do so consistently.
A second hypothesis was that the items would not have parallel error distributions. Because the wording of each of the items is different and because each was reduced to a dichotomous variable, it was expected that the false positive and false negative rates would vary by item. Overall, this hypothesis was also supported by the best fitting LCA model. Each indicator had higher false positive rates than false negative rates, where the false positive rates were in the range of 3 to 4 percent. It was found that question 62 (i.e., Did you drink alcohol or use drugs before you had sexual intercourse the last time?) had the highest potential for overreporting, approximately 4.5 percent (which is unweighted). Although some of this may be balanced by the false negative rate, there still may be potential for bias. One rationale for this finding may be that the question asks about two sensitive behaviors, sexual activity and alcohol/drug use, whereas the other items refer only to sexual activity.
As mentioned, LCA can be used to determine whether an indicator is an accurate measure of the true score. The YRBS relies on a single item to estimate the prevalence of sexual activity among youth. In practice, the estimates exclude any responses that are inconsistent with subsequent items. For example, if a respondent indicates he or she has never had sexual intercourse in question 58, but responds to the next question (i.e., question 59) that he or she has had two sexual partners, the data are excluded from the estimates. The accepted LCM showed that if these responses were included in the estimates of sexual activity, the estimates may be biased toward overreporting.
The analyses presented in this paper would have benefited greatly by incorporating weighting adjustments to the data. Patterson, Dayton, and Graubard (2002) showed that weighting adjustments can be used in conjunction with a latent class analysis to provide more accurate standard errors. However, when analyzing the YRBS data the weights may reduce the amount of sparseness found in the crossclassification table. In addition, using the weighted data (along with, perhaps a greater sample size) may allow for the inclusion of more indicator variables and/or other grouping variables in the LCM. There was some evidence that the error rates for males and females varied across indicators. When the grouping variable for sex was included in the model, the local independence assumption was rejected. Unfortunately, subsequent models testing for gender differences among the indicators were not possible due to a lack of statistical power. Perhaps a weighted dataset containing more respondents would yield more significant results.
References
Biemer, P. P, & Wiesen, C. (2002). Measurement Error Evaluation of SelfReported Drug Use: A latent class analysis of the U.S. National Household Survey on Drug Abuse. Journal of the Royal Statistical Society, A, 165(1): 97119.
Converse, J. M., & Presser, S. (1986). Survey Questions: Handcrafting the standardized questionnaire. Thousand Oaks, CA: Sage Publications, Inc.
Fowler, F. J. (1995). Improving Survey Questions: Design and evaluation. Thousand Oaks, CA: Sage Publications, Inc.
Groves, R. M. (1989). Survey Errors and Survey Costs. New York: Wiley.
Hagenaars, J. A. (1988). Latent Structure Models with Direct Effects Between Indicators: Local dependence models. Sociological Methods and Research, 16: 379405.
Hagenaars, J. A. (1990). Categorical Longitudinal Data: Loglinear panel, trend, and cohort analysis. Newbury Park, CA: Sage Publications.
Hagenaars, J. A. (1993). Loglinear Models with Latent Variables. Thousand Oaks, CA: Sage.
Knoke, D., & Burke, P. J. (1980). Loglinear Models. Beverly Hills, CA: Sage Publications.
Lin, T. H., & Dayton, C. M. (1997). Model Selection Information Criteria for NonNested Latent Class Models. Journal of Educational and Behavioral Statistics, 22(3): 249264.
McCutcheon, A. L. (1987). Latent Class Analysis. Thousand Oaks, CA: Sage Publications, Inc.
Patterson, B. H., Dayton, C. M., & Graubard, B. I. (2002). Latent Class Analysis of Complex Sample Survey Data: Application to dietary data. Journal of the American Statistical Association, 97(459): 721741.
Poon, W., Tang, M., Wang, S. (2003). Influence Measures in Contingency Tables With Application in Sampling Zeros. Sociological Methods and Research, 31(4): 439452.
Reiser, M., & Lin, Y. (1999). A GoodnessOfFit Test for the Latent Class Model When Expected Frequencies Are Small. Sociological Methodology, 29: 81111.
Tourangeau, R., & Smith, T. W. (1996). Asking Sensitive Questions: The impact of data collection mode, question format, and question context. Public Opinion Quarterly, 60: 275304.
Vermunt, J. K. (1997). LEM: A general program for the analysis of categorical data, users manual. Tilburg, The Netherlands: Tilburg University.
Appendix A
Q58.Have you ever had sexual intercourse?AYesBNoQ59.How old were you when you had sexual intercourse for the first time?AI have never had sexual intercourseB11 years old or youngerC12 years oldD13 years oldE14 years oldF15 years oldG16 years oldH17 years old or olderQ60.During your life, with how many people have you had sexual intercourse?AI have never had sexual intercourseB1 personC2 peopleD3 peopleE4 peopleF5 peopleG6 or more peopleQ61.During the past 3 months, with how many people did you have sexual intercourse?AI have never had sexual intercourseBI have had sexual intercourse, but not during the past 3 monthsC1 personD2 peopleE3 peopleF4 peopleG5 peopleH6 or more peopleQ62.Did you drink alcohol or use drugs before you had sexual intercourse the last time?AI have never had sexual intercourseBYesCNoQ63.The last time you had sexual intercourse, did you or your partner use a condom?AI have never had sexual intercourseBYesCNoQ64.The last time you had sexual intercourse, what one method did you or your partner use to prevent pregnancy? (Select only one response.)AI have never had sexual intercourseBNo method was used to prevent pregnancyCBirth control pillsDCondomsEDepoProvera (injectable birth control)FWithdrawalGSome other methodHNot sure
The NHSDA changed to computerassisted selfinterview following the 1999 collection cycle.
M[j ! 1 L S T c k p z (
)
*
,
7
8
:
E
M
_
h
i
s
/1SikȰȰhm_CJaJh)s6CJaJhZCJaJh%CJaJhCJaJhPCJaJhlCJaJh%CJaJhuSCJaJh:CJaJh8@CJaJ@
LM[jpq$dh`a$gd
$dha$gd%
$dha$gd8@
$dha$gd%i['@AEQ[ehG
H
V
W
Y
Z
^
e
h
q
~
./9GJK/ڷگҧʯʟʗҗʗʏʏʏʇʯhm_CJaJhVgCJaJh%CJaJhauCJaJhyCJaJhZCJaJhauhVgCJaJh)s6CJaJhCJaJhlCJaJh%CJaJhPCJaJhPhPCJaJhPh%CJaJ5Hoqr+,GH"ͼ{{{{{{hZhZ6CJaJh
CJH*aJhZhZCJH*aJ!jZhZhZCJEHUaJjQC
hZCJUVaJ!jhZhZCJEHUaJjC
hZCJUVaJjhZCJUaJh%CJaJh
CJaJhZCJaJ0q %
%#%((G(H(++++++
$dha$gd
$dha$gdV$dh`a$gdl
$dha$gd%
$dha$gdZ$dh`a$gd$dh`a$gdZ"&*,06268:<@RVZ\bT뺲몞{skkkhSACJaJhlCJaJh
hr/CJH*aJhZhr/CJH*aJhr/hr/CJaJhr/hr/CJH*aJhr/CJaJh%CJaJh
CJaJhZhZCJaJhZCJH*aJhZ6CJaJhZhZCJH*aJhZCJaJhZhZ6CJaJ*,CRlzlt~2s9X`wz79.yͽŭŵŵŵŵjh%0JCJUaJheh%CJaJh)s6CJaJh%CJaJheCJaJhMuCJaJhVgCJaJhSACJaJhY,h
CJaJhlCJaJh%CJaJh
CJaJ:"3M~ . 0 I !!"""""#:#=#@#I####$!$z$$$$$$%%
%%%ЬЬЬЬдȴشش褘h h 6CJaJh he6CJaJheCJaJh&CJaJh/CJaJ*h
h
CJaJhm_CJaJhd NCJaJh)s6CJaJh%CJaJh
CJaJhMuCJaJhlCJaJ7%%%%"%#%<%H%M%q%u%{%~%%%%%%%%%%%A&K&&&&&&')'f'y'''''''''''''ׯׯקח}h%6CJaJhU^6CJaJhU^CJaJhRCJaJhlCJaJhhCJaJh)s6CJaJh%CJaJhMuCJaJhVgCJaJh CJaJhlCJaJheCJaJh h 6CJaJh he6CJaJ''
((((($(%(&('(/(0(C(D(E(F(G(H(I(\(](^(_((((Ͳ͒ymy^MmyDyhV6CJaJ!j
hVhVCJEHUaJjC
hVCJUVaJjhVCJUaJhVCJaJ!jehU^hVCJEHUaJjEC
hVCJUVaJ!jhU^hVCJEHUaJjC
hVCJUVaJjhU^CJUaJhU^CJaJh%heCJaJhlh%CJaJhlh)s6CJaJh)s66CJaJ(())
)%)*);)<)O)P)Q)R){))))))))))**E*F*J*K*******6+Ƕsb֮!j*hVhTCJEHUaJjC
hTCJUVaJ!jhVhVCJEHUaJjTC
hVCJUVaJh)s6h)s66CJaJh)s6CJaJ!jhVhVCJEHUaJjC
hVCJUVaJjhVCJUaJhhT6CJaJhVCJaJhV6CJaJ#6++++++++++++++++;,I,M,,,,,+0;=MTxxphhph`p`h_>CJaJh%CJaJhCJaJh)s6CJaJhVCJaJhkCJaJhh6CJaJhlCJaJ!jhhCJEHUaJjC
hCJUVaJ!jhhCJEHUaJjoC
hCJUVaJjhCJUaJhCJaJhCJaJ%++,,22`3333*5+5G5H5999
$dha$gdX
$dha$gdVg
$dha$gdVg$dh`a$gdldh^gdi5Ldh]Lgdi5$dh`a$gd_f$dh`a$gdk
$dha$gdVv.z.............C/D/Z/a/////00<0=00000000000 1
111N1Q1r1s1x1y11111111111!2"2C2D2G2пппзЫЫЫЫЫЫЫЫЫЫЫhm_6CJaJhq}Zhq}Z6CJaJhQCJaJhi5CJaJhq}Z6CJaJhq}ZCJaJhY,CJaJhkCJaJhS/CJaJh)s6CJaJhCJaJ=G2O2[2c222222233333:3;3<3=3>3\3]3^3_3`3j3l3p3y3z33پpg[O[ghi5hY,5CJaJhY,hY,5CJaJh$%5CJaJhY,h#5CJaJh8
CJaJjh$%UjbD
h$%CJUVaJh"jh"Ujhi5Uj9D
hi5UVhi5jhi5Ujh"UjC
h_fUVjh_fUh_fhq}ZhkCJaJhq}ZCJaJhm_CJaJ33344444)5*5+5,5C5D5E5F5G5H566666666666l7p777777778ӿ~r~r~i~~r~r~r~r~ahCJaJhVgCJH*aJhV;hV;6CJaJhV;CJaJhV;CJH*aJhVgCJaJ!j+#hRhRCJEHUaJjC
hRCJUVaJhRCJaJjhRCJUaJh8
CJaJhCJaJhS/CJaJhq}ZCJaJh#CJaJhY,hq}Z5CJaJ&88*8+8,88\8`8a8x8y8z8{888889 99999x9y99999ûû{{{{skbkVhhX6CJaJhX6CJaJhXCJaJhhCJaJhhVg6CJaJhVgCJaJ!j(hhCJEHUaJjsC
hCJUVaJjhCJUaJhCJaJhi5CJaJ!ja&hM2hM2CJEHUaJjD
hM2CJUVaJhM2CJaJjhM2CJUaJ99999999999999999999::::+:,::.:0:b:ºއ{l[{Sh)s6CJaJ!j/hXhXCJEHUaJjhC
hXCJUVaJjhXCJUaJhlCJaJhVghhCJaJ!jU+hhfCJEHUaJjC
hfCJUVaJhCJaJjhCJUaJhVgCJaJhhCJaJhXCJaJhhX6CJaJhhX6CJH*aJ9::/:0::::=}=??DDGFHFIFQFRF
$dha$gdqy9
$dha$gdY,$dh`a$gdLa$dh`a$gdx
$dha$gd%$dh`a$gdl
$dha$gd/b:::::::
;;;;;;;$;%;4;;;;;9<:<Q<e<h<q<r<<<<<<<<d=i={==}===???"@иȰШȨШаИАzhY,CJaJjhLa0JCJUaJhLaCJaJh2CJaJh%CJaJhxCJaJh)s6CJaJhi5CJaJhW6
CJaJhauCJaJhdCJaJh#h%6CJaJh#hau6CJaJhXCJaJh/CJaJ"@N@Q@`@i@r@u@{@@@@@@@@@@@@@@AAAA%AAAAAAAA B'B(BBBBBBBC/C1C3CICMCCCCDDDD)D7DVDآؚؚh_>CJaJhi5CJaJh&CJaJhS/CJaJjhS/0JCJUaJh)s6CJaJhbCJaJhm_CJaJhxCJaJh`CJaJhauCJaJhdCJaJhxCJaJ9VD]DDDUEEEEFFFFF FFFFFF+F6FFFGFHFIFPFQFRFXF]FdFFFFFFFȺvmveehZCJaJhZ6CJaJhY,hY,6CJaJhCJaJhCJaJhqy9CJaJh_>CJaJhxCJaJhY,CJaJhS/CJaJjh#30JCJUaJhTCJaJhCJaJh#3CJaJhbCJaJh=CJaJhm_CJaJh&CJaJ$RFdFUIVIXKYKKKKKKKKKK$$Ifa$gd$m $Ifgd$m
$dha$gdxH$dh`a$gdY,
$dha$gdY,FUGYG]GGGGGGGGGGGGGGGXHrHvHyH~HHHHHHHIIIII#IQITImIvIwI}IIII`JWKYKhKmKKKKȸаذظШиШиМh)CJaJhxH5CJaJhZ5CJaJhL!hxH5CJaJh0CJaJh0CJaJhLaCJaJhZCJaJh]CJaJhbCJaJh/{~CJaJhM2CJaJhY,CJaJh#3CJaJhSCJaJ3KKkd3$$Iflִv=
Y B"
t06 44
laKKKKKKKKK$$Ifa$gd$m $Ifgd$mKKkd3$$Iflִv=
Y B"
t06 44
laKKKKKKKKK$$Ifa$gd$m $Ifgd$mKKkdb4$$Iflִv=
Y B"
t06 44
laKKKKKKKKK$$Ifa$gd$m $Ifgd$mKKkd
5$$Iflִv=
Y B"
t06 44
laKKKKKKKKK$$Ifa$gd$m $Ifgd$mKKkd5$$Iflִv=
Y B"
t06 44
laKKKL LLLLL$$Ifa$gd$m $Ifgd$mLLkdc6$$Iflִv=
Y B"
t06 44
laLLL"L'L,L1L5L6L$$Ifa$gd$m $Ifgd$m6L7Lkd7$$Iflִv=
Y B"
t06 44
la7L:L?LDLILNLSLXL\L$$Ifa$gd$m $Ifgd$m\L]Lkd7$$Iflִv=
Y B"
t06 44
laK]LLLLLLMMMMMMNQN[NNNNNNNNNNOO.O/O2OOOP
PP"P'P/P7PNPPPRPSPXP]PvPPQ Q=Q?QIQQQǿϷϷןןϟϧןhPCEhPCECJaJhXjCJaJh62CJaJh rCJaJhXDCJaJh
CJaJh$7,CJaJhqTCJaJh0CJaJhLaCJaJhLah`
UCJaJhLahLaCJaJh]hKCJaJ5]LLLQPRPQQQQQ
RRR$$Ifa$gdd $Ifgddgd$dh`a$gd r
$dha$gdqy9$a$gdLaQQQQQQRSSSS/S1S6SSSGTUCUUKVLVMVOVVVVY[[[ŽzrzreWrzrzOhDTCJaJjhLa0JCJUaJhLahLa0JCJaJhLaCJaJhz{CJaJh`hbCJaJhXCJaJhbCJaJhY,hY,6CJaJh r6CJaJh62h62CJaJh`
UCJaJh62CJaJhRehReCJaJhRe5CJaJhh5CJaJh05CJaJhK5CJaJRR*R4RdEdddddd*e`eee
ff"f*f.f0f;fKfQfRffffffffff5g6g;gCJH*aJh>6CJaJh>CJaJh#+CJaJhCJaJhCJaJhCJaJhCJaJjh&0JCJUaJ&mknkuk{kkkk
llllllllllllll4m7mmnnnnnnp^pbp^qrrrǿtld\\ddhtCJaJh$uCJaJh&CJaJjh0JCJUaJhCCJaJh[CJaJhLahCJaJh6CJaJhP6CJaJh5dCJaJhCJaJhPCJaJhCJaJh&h5d5CJaJh&hd N5CJaJhQ{5CJaJhQ{hd N5CJaJh5d$nkkknnrruuu0u3uHuKuSuWu[u$$Ifa$gdFD $IfgdFDgdFD
$dha$gdqy9$h^h`a$gd5drr#r2r3rrrsssssssssstt
tt8t=tttttttttuuuu
uu,u˸˧˟˗˸ˏˏˇ{r{gh%hFDCJaJh5CJaJhhFD5CJaJh=CJaJh)s6CJaJhCJaJh$uCJaJhCJaJh#+6CJaJh#+CJaJh[h[CJaJhFDCJaJhFDh[6CJaJhCCJaJhtCJaJh#+5CJaJh[CJaJ%,u/uIuJuousuuuvuwuyuzuuuuuuuuuuuuuvvvv6v8v9v;vCJaJh1`CJaJhOCJaJhgCJaJhCJaJh)K#CJaJhOKCJaJhCJaJhM2CJaJh_sCJaJhCJaJhGCJaJh._h._CJaJh$uCJaJhh`CJaJ,zzzzzqbbb$dh$Ifa$gd`kdPk$$IflF,"
t0644
lazzzzzqbbb$dh$Ifa$gd`kdk$$IflF,"
t0644
lazzzzz/0deqbWLLLLL
$dha$gd%
$dha$gd$u$dh`a$gd._kdhl$$IflF,"
t0644
laefӂԂ˃Zxք'JVWXYo2@YZʍ/ذ萈xoxh&2>6CJaJh&2>CJaJh#CJaJhxCJaJh6CJaJh,lCJaJh_5CJaJh= CJaJhdCJaJh,CJaJh_sCJaJhgCJaJhOKCJaJhCJaJhOCJaJh+=CJaJh1CJaJh^CJaJ)eXY!"WX=gd&2>h^h`gd&2>dhgd&2>
$dha$gd&2>$dh`a$gdx
$dha$gd%/m
&EVWʏ_ǐ<=> 78Lْ때uj_때hGh&2>CJaJhh&2>CJaJ*huSh&2>CJaJh+=h+=CJaJh+=6CJaJh+=CJaJh`h&2>6CJaJh`h&2>CJaJhxJhh&2>CJaJhxJhh&2>6CJaJh,h&2>CJaJ*h`h&2>CJaJh&2>6CJaJh&2>CJaJhGh&2>6CJaJ%=>KLÓ^_ߕ $IfgdQdhgd.
$dha$gd&2>gd&2>h^h`gd&2>8T]^
ޕߕ;<?@Ǘژۘܘݘ+徶hS,hQCJaJhS,hQ6CJaJhS,h.6CJaJhS,h.CJaJhECJaJh.CJaJhTCJaJh&2>h&2>CJaJhY~)h&2>CJaJh&2>6CJaJh&2>CJaJh+=CJaJh+=h+=CJaJ2ߕ{o{$$Ifa$gdQ $IfgdQzkdl$$Ifl0,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdam$$IflF,"
t0644
laqhh $IfgdQkdm$$IflF,"
t0644
la<{{ $IfgdQzkdgn$$Ifl0,"
t0644
la<=>@d{o{$$Ifa$gdQ $IfgdQzkdn$$Ifl0,"
t0644
ladefhqh\h$$Ifa$gdQ $IfgdQkdAo$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdo$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdGp$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdp$$IflF,"
t0644
laĖqh\h$$Ifa$gdQ $IfgdQkdMq$$IflF,"
t0644
laĖŖƖȖՖqh\h$$Ifa$gdQ $IfgdQkdq$$IflF,"
t0644
laՖ֖זٖqh\h$$Ifa$gdQ $IfgdQkdSr$$IflF,"
t0644
laqhh $IfgdQkdr$$IflF,"
t0644
la@{{ $IfgdQzkdYs$$Ifl0,"
t0644
la@ABDh{o{$$Ifa$gdQ $IfgdQzkds$$Ifl0,"
t0644
lahijluqh\h$$Ifa$gdQ $IfgdQkd3t$$IflF,"
t0644
lauvwyqh\h$$Ifa$gdQ $IfgdQkdt$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkd9u$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdu$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkd?v$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdv$$IflF,"
t0644
laqhh $IfgdQkdEw$$IflF,"
t0644
laǗ{{ $IfgdQzkdw$$Ifl0,"
t0644
la?{o{$$Ifa$gdQ $IfgdQzkd5x$$Ifl0,"
t0644
la?@ACqh\h$$Ifa$gdQ $IfgdQkdx$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkd%y$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdy$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkd+z$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdz$$IflF,"
t0644
laĘqh\h$$Ifa$gdQ $IfgdQkd1{$$IflF,"
t0644
laĘŘƘȘ٘qh\h$$Ifa$gdQ $IfgdQkd{$$IflF,"
t0644
la٘ژۘܘqhh $IfgdQkd7$$IflF,"
t0644
laܘݘ6{{ $IfgdQzkd$$Ifl0,"
t0644
la+56nopqvzřƙ
58_pefghijƛǛEF*+Z[ٶٶٮh<hLaUjhLa0JUhjh0JUhh$%hS,CJaJhS,h.6CJaJhS,hQ6CJaJhS,hQCJaJhS,h.CJaJhS,hQ56CJaJhS,h.56CJaJ/678:^{o{$$Ifa$gdQ $IfgdQzkd'}$$Ifl0,"
t0644
la^_`bfqh\h$$Ifa$gdQ $IfgdQkd}$$IflF,"
t0644
lafghjmqh\h$$Ifa$gdQ $IfgdQkd~$$IflF,"
t0644
lamnopqhh $IfgdQkd~$$IflF,"
t0644
lapqvƙ{{ $IfgdQzkd$$Ifl0,"
t0644
laƙǙșʙ{o{$$Ifa$gdQ $IfgdQzkd$$Ifl0,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkd$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdz$$IflF,"
t0644
laqhh $IfgdQkd$$IflF,"
t0644
la{{ $IfgdQzkd$$Ifl0,"
t0644
la{o{$$Ifa$gdQ $IfgdQzkd$$Ifl0,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkdZ$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkd݂$$IflF,"
t0644
laqh\h$$Ifa$gdQ $IfgdQkd`$$IflF,"
t0644
la
2qh\h$$Ifa$gdQ $IfgdQkd$$IflF,"
t0644
la2346Aqh\h$$Ifa$gdQ $IfgdQkdf$$IflF,"
t0644
laABCEWqh\h$$Ifa$gdQ $IfgdQkd$$IflF,"
t0644
laWXY[dqh\h$$Ifa$gdQ $IfgdQkdl$$IflF,"
t0644
ladefghiƛEqiaQLJEgdLagd5d$h^h`a$gd_f$a$gd$%dhgd.kd$$IflF,"
t0644
la Because of the sensitive nature of the survey, the raw data were provided for the present study with the understanding that the four states and city would not be identified. Additionally, these data were provided prior to any standard analyses and therefore no descriptive statistics such as response rates were available.
Although the responses to these items have been dichotomized in such a way as to render them replicate measures of sexual activity, this was not the intent of the original survey items.
The sample size includes missing data.
In fact, the LCA models were tested while including question 61, which resulted in boundary estimates. These issues are discussed further below.
Adding a grouping variable also increases the degrees of freedom.
The maximum likelihood estimate reaches a value of 0 or 1 before the model converges.
Replicate Measures PAGE 17
PAGE
E*Z[~h]hgd8@h]hgd&`#$gd$a$gd8@gdLa[rsyz}~ĺİĬh$%hpLh<
h0Jjh0JUhh<0JCJaJmHnHuh8@h0JCJaJ!jh8@h0JCJUaJh8@hCJaJgd5d
)
01h:p%/ =!"#$%ZDd
hb
c$A??3"`?2%R`kur 89D+`!x%R`kur 89 @`Fxcdd``vbd``baV d,FYzP1n:&B@?b =
UXRY7&meabMVKWMc~x V8pNHq=`l9)3`91Ff0?7bc?ld670zLȆ@l,0 Ma(wZsqF=Ab,4sùu\6pfmSFm=TĤ\Y\q1(2tAB;/G3XHyumDd
@hb
c$A??3"`?2H6BRaWe+`!H6BRaWe
@pYxڝR=KA)D(b/(hRX3BuW?FӅOӟI
zOyVc6x,}w}ƙsDVRơ`zuxFg].#R;c%Wk4:;:&e5i#!s}F8dUpsFw/HN
Dd
lb
c$A??3"`?2Cu!;$E_8=+`!Cu!;$E_8=^@Xhxcdd``^ @c112BYL%bpu1z/7= {W wq~N_jzhiOZ{̘1s"d_"%}^1/Fxԭ&oNz!!^?
tQy~zW7A^Nlxw8}J@eͰ>{gW{mhAzYF_߽~ty닇t)Ѽk Cu
ۺOq9ηLn/S}pz;jۨmcr{/g6K+Dd
b
c$A??3"`?2u,q՜5^imQ
+`!I,q՜5^imۦ@2hxmP=KA}3w~dȑJLa[ZSL̙R[A8 G\g.bo7
}`ņdJDlUtH;%Wx6ubR撹ZjjMoYzqhu<8(
qEK{
]KĲ
}re[cǃG'Sa?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~lRoot Entryp F 8qFData
rWordDocumento+ObjectPoolr =_q 8q_1139461403F =_q aqOle
CompObjfObjInfo"%(+.1458;<=>?BEFGHINOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstz}~
FMicrosoft Equation 3.0DS EquationEquation.39qV
yikm
=Xik
+im
FMicrosoft Equation 3.0DS EqEquation Native r_1139461457MF aq aqOle
CompObj
fuationEquation.39q^$
yikm'
=Xik
+im'
FMicrosoft Equation 3.0DS EquationEquation.39qObjInfo
Equation Native
z_1139468704 'F aq aqOle
CompObj
fObjInfoEquation Native _1139468101F aq dq
ijklABCD
=ijkltABCDXt=1T
"
FMicrosoft Equation 3.0DS EquationEquation.39q1
ijkltOle
CompObjfObjInfoEquation Native MABCDX
=ijklt2A2B2C2DX
=tX
it2AX
jt2BX
kt2CX
lt2DX
FMicrosoft Equation 3.0DS EquationEquation.39q_1139468716F dq dqOle
CompObjfObjInfo 7Hj<
ijklABCD
FMicrosoft Equation 3.0DS EquationEquation.39q?
ijkltEquation Native !S_1139468820"F dq dqOle
#CompObj $fObjInfo!&Equation Native '[_1139468884$F dq dqOle
)ABCDX
FMicrosoft Equation 3.0DS EquationEquation.39q+(4
it2AXCompObj#%*fObjInfo&,Equation Native G_1139469029\)F dqifqOle
/CompObj(*0fObjInfo+2Equation Native 3
FMicrosoft Equation 3.0DS EquationEquation.39qwx
tX
=it2AX
=1i
"i
"
FMicrosoft Equation 3.0DS Eq_1139469679.FifqifqOle
6CompObj/7fObjInfo09$#
!"H%&'(j*+,./0123456789:;<=>?@ABCDEFGiJKLMNOPQRSTUVWXYZ[\]^_`abcdefghkmnpoqrtsuvw~xyz{}uationEquation.39q6h
FijkltABCDX
=iA
jB
kC
kD
tX
itAX
jtBX
ktCX
ltDXEquation Native :R_1139469818,63FifqifqOle
@CompObj24Af
FMicrosoft Equation 3.0DS EquationEquation.39qZ\D
GijkltABCDX
=+iA
+jB
+kC
+kD
+tX
+itObjInfo5CEquation Native Dv_1139917008k8 gyJthqhqOle
JAX
+jtBX
+ktCX
+ltDXS ObjInfo79KContents6_1144644759Hf< gyJthqhqOle
LABC7CyHdArial#Arial6)
vXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)('& %'$#"! ?>=<;:9876543210O#N M&LK JIH G#FED C B A@+_^]\[ZYXW*VU TSR Q#Ponm'lkjihgfedcba`!~}{zyxw#vutsrqp!!,,!!*,,!!!%%% , ###### 'Arial #Arial2%
c`` '
/
.

,
+*
)
(
'&%!$#"
!
?>=<;
:
9876543210ONMLKJIHGFEDCBA@%_^]
\
[
ZYXW!VUTSRQPonmlkjihgf
edcba`
~}
{
zyxw!vut
sr
qp%
%
%
!
%
%
!!
%!Arial#Arial4'
wXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)(' &%'$#"!
?>=<;:9876543210O#N M%LKJIH G#FED C BA@_^]\[ZYXW,VU TSR Q#Ponm&l
kj
i
hgf
edcba`!~}{zyxwvutsrqp!!,,,
!!*,
,
!!!
%%% , ###### '
vjNn*Uj=jiCenturyK
Cyrillicpjj
8Arial~(?NormalArial
Default Shape( Default Line
D0!?0?A
B
11NLayer 1A 9 2
( ?@?p@?PPPP0n1oq/o2PP U[ज़?@ज़?p@?ज़?@ज़?p@? m
x
x
!
~=~=CfC/ C12&"vm1jXkdjj*ii(7R"3456 X
Connector 6
( w?@` z??j@jj5O>>>5>O `?@?ӿ?`?@?ӿ? s@@@
@
@!
={X=ADY/ C12&"e1jXkdj`jj,iB7N"3456 AProcess 7
( ? @+z?g@?j@jj5O>>>5>O ?@?=g@??@?=g@? s@@@
@
@!
7=vX=D'C/ C12&"e1jXkdjjiiA7N"3456 B Process 8
( w?@` z?w@?j@jj5O>>>5>O `?@??w@?`?@??w@? s@@@
@
@!
={X=ADfzC/ C12&"e1jXkdj0jj4kA7N"3456 C Process 9
( @o{?@`o{?@?j@jj5O>>>5>O ?@?O@??@?O@? s@@@
@
@!
zۓ=CX=D D/ C12&"e1jXkdjIkpjd;iA7N"3456 D Process00}hz000? C00^h000? C00^h000? C00?h=000? C C ( Cj<j (Cj<j (jAwR (jAwR (Cj<j (Cj<j (jAwR (0jAwR&,!h?,PRINT;=M ObjInfo>uContents)<_1141637218a:A gyJthq$kq#Arial ~'
~
~ ~0 2
OMX $RRR 2
.A $RFRFR 2
B $RRR 2
C $RIIRRI 2
kD v6$R>h.RRh>.Rj3$R6/RR6/Rj3$R/6RR/6RKZ7$RI.M?hRIIRM.h?IR$$"System
& ~ABC7CyHdArial#Arial6)
vXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)('& %'$#"! ?>=<;:9876543210O#N M&LK JIH G#FED C B A@+_^]\[ZYXW*VU TSR Q#Ponm'lkjihgfedcba`!~}{zyxw#vutsrqp!!,,!!*,,!!!%%% , ###### 'Arial#Arial4'
wXX /!!!!!!!!! !
!!!
!!!!!!!!!!!!!!!!!!! !
"#$%'&' ()*+,./0123456789:;<=>?@ABC D EFG#H IJKLM%N O#PQ#R STU VW,XYZ[\]^_`abcdef
ghi
j
kl
m&nopqrstuvwxyz{}~!!
,,,!!!
,
,*!
!!%%%, ###### 'Arial #Arial* _XX /
!"#$%&'()*+,
.
/
0123456789:
;
<=>?@$ABCDEFGHI
JKLMNOPQRSTUVW#XYZ[
\
]
^_`abcdef
ghijklmnopqrst
uvwxyz{ }~#$###!
#
vN*K
8Arial~?NormalArial
Default Shape(
Default Shape Default Line Default Line
?A
B
Ƚk1Layer 1 A 9 2
( ?@?@?PPPP0n1oq/o2PP U[ज़?@ज़?@?ज़?@ज़?@? m
x
x
!
~=~=A.DD/ C12&"vm1jXkdkHll$l(7R"3456X
Connector 6
( w?@` z?H@?j@jj5O>>>5>O `?@?F@?`?@?F@? s@@@
@
@!
={X=AD^D/ C12&"e1jXkdk$ll$lB7N"3456AProcess 7
( ? @+z?D@?j@jj5O>>>5>O ?@?'E@??@?'E@? s@@@
@
@!
7=vX=D>D/ C12&"e1jXkdk)l,l)lA7N"3456B Process 8
( w?@` z?@@?j@jj5O>>>5>O `?@?>@?`?@?>@? s@@@
@
@!
={X=ADD/ C12&"e1jXkdk/l*l.lA7N"3456C Process 9
( @o{?@`o{?@<@?j@jj5O>>>5>O ?@?';@??@?';@? s@@@
@
@!
zۓ=CX=DhD/ C12&"e1jXkdk4l46l3lA7N"3456D Process00{'000? C00'M000? C00'000?
C00{'000? C C0000
Q0000? C C (pkAwR (pkAwR (0jAwR (jAwR (
Cj<j (
Cj<j ( jAwR ( jAwR (Cj<j (Cj<j&,$`R,
A7O[JUChart%u\zm/Lgn}{>P@a@ja'&(33[gf/l`kj/nz{`bn{flnccv/hjaj}n{jk/mv/Ic`xLgn}{j}!//K`/A`{/]jb`yj11yj}/2/Ncfln{f`a!Yj}f`abne`}/2/Mnfl!Cji{/'yj}#/=&bfa`}/2/Mnfl!]fhg{/'yj}#/=&Fi/'bne`}/2/8!/Nak/'bfa`}/2/!?/`}/bfa`}/2/!>&&/[gjaNcfln{f`a!BhM`w/[gf/ifcj/xn/nyjk/xf{g/n/ajxj}/yj}f`a/`i/Ic`xLgn}{j}!//Fi/v`z//nyj/f{/fa/{gf/yj}f`a#/v`z/bnv/c`j/kn{n!//_cjnj/l`a{nl{/v`z}/aj{x`}d/nkbfaf{}n{`}/`}/Bfl}`h}niw/{`/`m{nfa/{gj/cn{j{/yj}f`a!//V`z/lna/ifak/fai`}bn{f`a/nm`z{/{gj/b`{/lz}}ja{/yj}f`a/`i/Ic`xLgn}{j}/n{/g{{5 xxx!bfl}`h}niw!l`b!Jak/Fi(33[gf/l`kj/nz{`bn{flnccv/hjaj}n{jk/mv/Ic`xLgn}{j}!//K`/A`{/]jb`yj11Jak/\zm
Ole
vObjInfo@BwContentsI>_1139137183EF$kq$kqABC⦀_lyHdArial#Arial4'
wXX /!!!!!!!!! !
!!!
!!!!!!!!!!!!!!!!!!! !
"#$%'&' ()*+,./0123456789:;<=>?@ABC D EFG#H IJKLM%N O#PQ#R STU VW,XYZ[\]^_`abcdef
ghi
j
kl
m&nopqrstuvwxyz{}~!!
,,,!!!
,
,*!
!!%%%, ###### 'Arial #Arial* _XX /
!"#$%&'()*+,
.
/
0123456789:
;
<=>?@$ABCDEFGHI
JKLMNOPQRSTUVW#XYZ[
\
]
^_`abcdef
ghijklmnopqrst
uvwxyz{ }~#$###!
#
Arial#Arial6)
vXX /!!!!!!!!! !
!!!
!!!!!!!!!!!!!!!!!!! !"#$%'& '()*+,./0123456789:;<=>?@+AB C D EFG#H IJK LM&N O#PQ#R STU VW*XYZ[\]^_`abcdefghijklm'nopqrstuvw#xyz{}~!!,,!!!,,*!!!%%%, ###### '
v,iN*jļjA,iK
%,iiihj
8Arial~(?NormalArial
Default Shape( Default Line Default Line
?A
B
l1Layer 1 A 9 2
( ?(@?q@?PPPP0n1oq/o2PP U[ज़?(@ज़?q@?ज़?(@ज़?q@? m
x
x
!
~=~= rC/ C12&"vm1jXkdmlll(7R"3456X
Connector 6
( w?@` z?(@?j@jj5O>>>5>O `?@?g'@?`?@?g'@? s@@@
@
@!
={X=D0/ C12&"e1jXkdm̙lllB7N"3456AProcess 7
( ?@@+z?`i@?j@jj5O>>>5>O ?@?>>5>O `?@?@x@?`?@?@x@? s@@@
@
@!
={X=DhzC/ C12&"e1jXkdmmldmA7N"3456C Process 9
( @o{?@`o{?H@?j@jj5O>>>5>O ?@?OF@??@?OF@? s@@@
@
@!
zۓ=CX=nDD/ C12&"e1jXkdmmmtmA7N"3456D Process00R000? C00jR000? C00jR000?
C00KRI000? C C0000000? C C`f?P@fP?@^@?f?P@fP?@^@? C12&"51jXkdmn+ln5"3456 `f?p@fP?M@?f?p@fP?M@? C12&"51jXkdm؍nnn5"3456 `f?P@fP?@^@?f?P@fP?@^@? C12&"51jXkdmňnn5"3456 (0jAwR (jAwR (
Cj<j (
Cj<j ( jAwR ( jAwR (Cj<j (Cj<j (kAwR (kAwR&'
hQ,Ole
xCompObjDFyjObjInfoG{Equation Native 
FMicrosoft Equation 2.0DS EquationEquation.COEE29q*C
Pxvs
()=pgs
+i=1k
P
pixgsg=1G
_1142787609JF$kqmqOle
CompObjIKfObjInfoL
FMicrosoft Equation 3.0DS EquationEquation.39qnS(
ixgsx=1m
"
=1
FMicrosoft Equation 2.0DS EquationEquation.COEE29qEquation Native o_1139137651CWOFmqmqOle
CompObjNPjObjInfoQEquation Native \_1139137702TFmqmqOle
*@C
g
pgs
=1
FMicrosoft Equation 2.0DS EquationEquation.COEE29q*ZC
ln2f hjkCompObjSUjObjInfoVEquation Native v_1139138152RYFmqmqmgs()=l+lhA
+ljB
+lkC
+lmD
+lgX
+lsN
+lhgAX
+ljgBX
+lkgCX
+lmgDX
+lgsXN
+lhsAN
+ljsBN
+lksCN
+lmsDN
+lhgsAXN
+ljgsBXN
+lkgsCXN
+lmgsDXN
FMicrosoft Equation 2.0DS EquationEquation.COEE29q*ZC
lhA
=lhgAX
=lhgAX
=lhsAN
=lhsAOle
CompObjXZjObjInfo[Equation Native vN
=lhgsAXN
=lhgsAXNs
h
=0s
h
g
h
h
K& #Arialw_11411980461?^ gyJtOrqtqOle
PRINT]_*
ObjInfo`h
۷ww w0 2
sX $gwwgg 2
KA $g&x&xgg& 2
eB $g@w@wgg@ 2
C $gZxZxggZ 2
D >K$gSCggSCgH$gKDggKDgH$gDwKggwDKgwL$gfCjTgffgjCTfgv$xpxxpx$wwww$xxxx $@ @@ @:$Z::ZZ::Z"System
&w"S ] R#Arialw
ww w0 2
IXContents>_1141636315c gyJttqtqOle
PRINTbd
TUABC7CyHdArial#Arial6)
vXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)('& %'$#"! ?>=<;:9876543210O#N M&LK JIH G#FED C B A@+_^]\[ZYXW*VU TSR Q#Ponm'lkjihgfedcba`!~}{zyxw#vutsrqp!!,,!!*,,!!!%%% , ###### 'Arial#Arial4'
wXX /!!!!!!!!! !
!!!
!!!!!!!!!!!!!!!!!!! !
"#$%'&' ()*+,./0123456789:;<=>?@ABC D EFG#H IJKLM%N O#PQ#R STU VW,XYZ[\]^_`abcdef
ghi
j
kl
m&nopqrstuvwxyz{}~!!
,,,!!!
,
,*!
!!%%%, ###### 'Arial #Arial* _XX /
!"#$%&'()*+,
.
/
0123456789:
;
<=>?@$ABCDEFGHI
JKLMNOPQRSTUVW#XYZ[
\
]
^_`abcdef
ghijklmnopqrst
uvwxyz{ }~#$###!
#
vN*K
8Arial~?NormalArial
Default Shape(
Default Shape Default Line Default Line
?A
B
ik1,K)Layer 1^ VA 9 2
( ?U@ ?{6@?PPPP0n1oq/o2PP U[v?U@v?0@?v?U@v?0@? m
x
x
!
== /A6C/ C12&"1jXkdD!kkDkkH::7q"3456X
Connector 6
( l@?F@C?J@?j@jj5O>>>5>O rh?@+?J@?rh?@+?J@? s@@@
@
@!
a= =cDA/ C12&"1jXkdD!kl4kdln447k"3456AProcess 7
( V?G@C?@[t@?j@jj5O>>>5>O x?@+?t@?x?@+?t@? s@@@
@
@!
==$cDvjC/ C12&"1jXkdD!klltlm447k"3456B Process 8
( l@?F@C?@[E@?j@jj5O>>>5>O rh?@+?H@?rh?@+?H@? s@@@
@
@!
a= =cDD/ C12&"1jXkdD!klllm447k"3456C Process 9
( E?F@E? @?j@jj5O>>>5>O x?@+?@?x?@+?@? s@@@
@
@!
,=pß=xcDXD/ C12&"1jXkdD!klllm447k"3456D Process00;g000? C00E9000? C00E9000?
C00;000? C C0000$000? C0000000? C0000000? C00::i000? C00::000? C C (C8kLj (C8kLj (8kAwR (8kAwR (8kLj (8kLj (8kLj (8kLj (8kAwR (8kAwR (Cj<j (Cj<j ( jAwR ( jAwR (
Cj<j (
Cj<j (jAwR (0jAwR&,$`R,
$QQ 2
.A $QQFF 2
B $QQ 2
C $IQIQI 2
kD u}$gtgtdy$uudy$uuEY$ItKgIIKtgI $''' 2
IZS dd2$Rd2T2tRddRT2t2dR"System
&wObjInfoeContents;_1144816895h gyJttq1wqOle
ABC7CyHdArial#Arial6)
vXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)('& %'$#"! ?>=<;:9876543210O#N M&LK JIH G#FED C B A@+_^]\[ZYXW*VU TSR Q#Ponm'lkjihgfedcba`!~}{zyxw#vutsrqp!!,,!!*,,!!!%%% , ###### 'Arial #Arial2%
c`` '
/
.

,
+*
)
(
'&%!$#"
!
?>=<;
:
9876543210ONMLKJIHGFEDCBA@%_^]
\
[
ZYXW!VUTSRQPonmlkjihgf
edcba`
~}
{
zyxw!vut
sr
qp%
%
%
!
%
%
!!
%!Arial#Arial4'
wXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)(' &%'$#"!
?>=<;:9876543210O#N M%LKJIH G#FED C BA@_^]\[ZYXW,VU TSR Q#Ponm&l
kj
i
hgf
edcba`!~}{zyxwvutsrqp!!,,,
!!*,
,
!!!
%%% , ###### '
v,iN*j̼jA,iK
%,iiipj
8Arial~(?NormalArial
Default Shape( Default Line
?A
B
1!]Layer 1 2
( ?@?p@?PPPP0n1oq/o2PP U[ज़?@ज़?p@?ज़?@ज़?p@? m
x
x
!
~=~=A.DD/ C12&"vm1jXkdjkjQk(7R"3456X
Connector 6
( w?@` z?P@?j@jj5O>>>5>O `?@?N@?`?@?N@? s@@@
@
@!
={X=ADD/ C12&"e1jXkdj k%kkB7N"3456AProcess 7
( ? @+z?L@?j@jj5O>>>5>O ?@?'M@??@?'M@? s@@@
@
@!
7=vX=D~D/ C12&"e1jXkdjk\jkA7N"3456B Process 8
( w?@` z?H@?j@jj5O>>>5>O `?@?F@?`?@?F@? s@@@
@
@!
={X=AD^D/ C12&"e1jXkdjk(k8kA7N"3456C Process 9
( @o{?@`o{?@D@?j@jj5O>>>5>O ?@?'C@??@?'C@? s@@@
@
@!
zۓ=CX=Dh>D/ C12&"e1jXkdj kkkA7N"3456D Process00{'000? C00'O000? C00'000? C00{'000? C 6
( w?y@` z?@?j@jj5O>>>5>O `?y@?Ϻ@?`?y@?Ϻ@? s@@@
@
@!
={X=CD/ C12&"e1jXkdj0lBkt0l`7N"3456SProcess0000T000? C C (kAwR (kAwR (0jAwR (jAwR (Cj<j (Cj<j (jAwR (jAwR (Cj<j (Cj<j&hS,PRINTgiObjInfojContentsjG_1139917312m gyJt1wq1wqK& D #Arial ~
~
~ ~0 2
kX $gwwgg 2
KA $g&x&xgg& 2
eB $g@w@wgg@ 2
C $gZxZxggZ 2
D G$gGGggGGgI$gCOggCOgO$gDYggDYgXS$gfGu`gffguG`fgv$xpxxpx$wwww$xxxx $ 2
KS $99G$g9G)GIg99g)GIG9gfGI$gSO8CVgSSg8OVCSgYN$gmXMDfgmmgMXfDmgnS$g`dGxggd`xGg $@ @@ @:$Z::ZZ::Z"System
& ~ABC7CyHdArial#Arial6)
vXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)('& %'$#"! ?>=<;:9876543210O#N M&LK JIH G#FED C B A@+_^]\[ZYXW*VU TSR Q#Ponm'lkjihgfedcba`!~}{zyxw#vutsrqp!!,,!!*,,!!!%%% , ###### 'Arial #Arial2%
c`` '
/
.

,
+*
)
(
'&%!$#"
!
?>=<;
:
9876543210ONMLKJIHGFEDCBA@%_^]
\
[
ZYXW!VUTSRQPonmlkjihgf
edcba`
~}
{
zyxw!vut
sr
qp%
%
%
!
%
%
!!
%!Arial#Arial4'
wXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)(' &%'$#"!
?>=<;:9876543210O#N M%LKJIH G#FED C BA@_^]\[ZYXW,VU TSR Q#Ponm&l
kj
i
hgf
edcba`!~}{zyxwvutsrqp!!,,,
!!*,
,
!!!
%%% , ###### '
v,iN*`jjA,iK
%,ii%Lj
8Arial~(?NormalArial
Default Shape( Default Line
?A
B
1Layer 1,# $# 2
( @?@@?͛@?PPPP0n1oq/o2PP U[v?@v?̛@?v?@v?̛@? m
x
x
!
==C
D/ C12&"1jXkd$jjhijH::7q"3456X
Connector 6
( l@?#w@C?x@?j@jj5O>>>5>O rh?x@+?x@?rh?x@+?x@? s@@@
@
@!
a= =wD?sD/ C12&"1jXkd$jjjjn447k"3456AProcess 7
( V?#w@C?V#@?j@jj5O>>>5>O x?x@+?$@?x?x@+?$@? s@@@
@
@!
==sDڜD/ C12&"1jXkd$jjjjm447k"3456B Process 8
( l@?#w@C?@?j@jj5O>>>5>O rh?x@+?@?rh?x@+?@? s@@@
@
@!
a= =wDXD/ C12&"1jXkd$j}kiim447k"3456C Process 9
( E?`#w@E?@?j@jj5O>>>5>O x?x@+?@?x?x@+?@? s@@@
@
@!
,=pß=>>5>O rh?@+?x@?rh?@+?x@? s@@@
@
@!
a= =\C?sD/ C12&"1jXkd$j\kkأk447k"3456SProcess00000? C00^000? C00I^6000? C00v^P000? C00^j000? C0000#000? C0000=000? C C (Ckj (Ckj (Ckj (Ckj (k (k (
8kLj (
8kLj (8kLj (8kLj (8kAwR (8kAwR (jAwR (jAwR ( Cj<j ( Cj<j (
jAwR (
0jAwR (C8kLj (C8kLj (8kAwR (8kAwR (!l!l (!l!l (C8kLj (C8kLj (kk (kk&7hx,Ole
ObjInfolnContents A1TableXOh+'0(4P\h
The Youth Risk Behavior Survey (YRBS) is sponsored by the CDC and is designed to monitor behaviors that are related to causes he Preferred CustomervrefrefNORMAL.DOTuCarl Setzer
!"#$%&'()*+,./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSVYZ[\]^_`abcdefghijklmnopqrstuvwxyz{}~ABC7CyHdArial#Arial6)
vXX /!!
!!!
! !!!!!!!!!!!!!!!!!!!!!!!!!!/.,+*)('& %'$#"! ?>=<;:9876543210O#N M&LK JIH G#FED C B A@+_^]\[ZYXW*VU TSR Q#Ponm'lkjihgfedcba`!~}{zyxw#vutsrqp!!,,!!*,,!!!%%% , ###### 'Arial#Arial4'
wXX /!!!!!!!!! !
!!!
!!!!!!!!!!!!!!!!!!! !
"#$%'&' ()*+,./0123456789:;<=>?@ABC D EFG#H IJKLM%N O#PQ#R STU VW,XYZ[\]^_`abcdef
ghi
j
kl
m&nopqrstuvwxyz{}~!!
,,,!!!
,
,*!
!!%%%, ###### 'Arial #Arial* _XX /
!"#$%&'()*+,
.
/
0123456789:
;
<=>?@$ABCDEFGHI
JKLMNOPQRSTUVW#XYZ[
\
]
^_`abcdef
ghijklmnopqrst
uvwxyz{ }~#$###!
#
vN*K
8Arial~?NormalArial
Default Shape(
Default Shape Default Line Default Line
&0?A
B
i2vLayer 1 A 9 2
( ?C@?@@?PPPP0n1oq/o2PP U[ज़?C@ज़?@@?ज़?C@ज़?@@? m
x
x
!
~=~=xD/
C12&"vm1jXkdL#iNk'kk(7R"3456X
Connector 6
( w?g@` z?@?j@jj5O>>>5>O `?h@?@?`?h@?@? s@@@
@
@!
={X=zD^D/ C12&"e1jXkdL#ikk`kB7N"3456AProcess 7
( ?@h@+z?@?j@jj5O>>>5>O ?h@?'@??h@?'@? s@@@
@
@!
7=vX=uD>D/ C12&"e1jXkdL#ik@klkA7N"3456B Process 8
( w?g@` z?@?j@jj5O>>>5>O `?h@?@?`?h@?@? s@@@
@
@!
={X=zDD/ C12&"e1jXkdL#illllA7N"3456C Process 9
( @o{?h@`o{?@@?j@jj5O>>>5>O ?l@?'@??l@?'@? s@@@
@
@!
zۓ=CX=nyDhD/ C12&"e1jXkdL#illlA7N"3456D Process00m000? C00m=000? C00m~000? C00m000? C C 1
( ?ؑ@?+@?PPPP0n1oq/o2PP U[ज़?ؑ@ज़?,@?ज़?ؑ@ज़?,@? m
x
x
!
~=~=D`D/ C12&"vm1jXkdL#i<jdlԪk(7R"3456 Y
Connector00{&000? C00v&~000? C00v&=000? C00&000? C C (kk (6kk (Ck`;i (%Ck`;i (Ck`;i (Ck`;i (Ck`;i (Ck`;i (Cj<j (
Cj<j (jAwR (
jAwR (Cj<j (
Cj<j (jAwR (
0jAwR&,$`R,mQ=KA}3FmI
3ZYiaRXz .?!YH@@T0I00b&X1eyY*`B9+,g,Ԝ״JQԣgJpk9Cl\ų7qy445͋8bIR?~?>SN\H{1bU:«QjݶZk
?CKR+ز!x˧Mq֝8>NOIv>{0#$ztɋZ﹎m?f7_b@.olT%]{G+^˱lAF`_&s.;5%ODd
xb
c$A??3"`?2(g&\.G 8K+`!g&\.G 8K^&hxڝ=KAgר^41Q!
Zؤ`Zpbbb邥 XX U,"ZlmDܙ\Pg߹p\rx9e53MyWCo@,t䪴*rOQ9>~XLPqh7~B+8 rL3tb)oJ#'~C?!'>3r\
zcJjgTp=݊g2͜>0>~rRʯr`!~ý楀w;9*u@}h4+ _+܊!_8%&Qt;Wډh+mC# ]Y\^7l!٫
Dd
pb
c$A??3"`? 2Tگy["8<80+`!(گy["8<8%X":hxڝKTQϽ;S3NMJ='l8Zj5*6M*s1b@$!k*@i̛+{>{8?>10z̎Ɏ%,!0$+'KȦlu7 C3k}P^s
+J7[BjZԝJ̬e܅n7a9,b_w\zxw=D[q2X;kbmuF.YNM"rω뛏!^&j/N+_(\Pcz+w9:yl`G)q%ƝmW}2gͷWظ:&a?s~98gӋ_DOy(y&op+KgsW{
>ҿAHmu`^z{3"9Dd
\
36
3A
?
2s4CʅTدy¯O+`!G4CʅTدy¯TA;$#!xڵKQgfgCULZ<B鱰H4Kx)(*WE@PJmۙ>Ӕv~;<"@"!9D?i*PuDIT!ql&%Z"4k3L7$4);wwH~t#.7YI8rKoV_AݾfDb&LQn_S>^'<Ȝk(7̺so#k͔X5c5S⿀=?˔ƕkla:dQ!8kDS?VܡM<2wGVI,׳ܮCf,XRHϴ.9T*֤*u)ٛ[/fߺst{Y>G>!:;XisJwwd Σj%8ķv*9$Cϸj%nKoUjzpk
Z"Q^Ua8Oͺp_q'Dd
)`0
#A2Ӣ&8W+`!wӢ&8W }"dExڵkQǿ3fBEoMB)U!^ě]T/9Vo^{$xzKBT\g۷&F{3}p Tyjs"N(&<(SuM5dFajԏhz_[U]}Tzs?EvJ,"[ɘJp!Ewpj1k<jFOtO#q:ޘU.?vf 4WX*jF۬ѧbߓV2kgahr1ٞ'ӌ#5/I,G4b~{Dtc/sƪL
dY&V\mfWDs?vM<"yrnZŻ=T@4xTmq!fvRnsWct w~;kv[骃TvK6eٯ76$˩at_n.<]CL9x:rи7b>s]蛾{ms]ފ8u$L{.qk$K7uL[XP9Dv*(.QtW?wO&IzZMK*pa'KVh=ؠT"8ӆÚUFsQS:
?̮]A>h3hb)T{\kdw;[(43ȷ{ѝgEMkgu[{ֹ(ɦ [6kt
m}^ЉT;T[npNw$kr9hig oXޡt?l6Dd
$ b
c$A??3"`?
2?'4.\\o#+`!T?'4.\*"xcdd``ef``baV d,F1J@Yp1B@?6 YzjxK2B*R>]k&gBܤr@;_ʣh .#9؈$ĘKɌ`# Pś47$37X/\!(?71!2bhWаPMCLf`RQV2.RI,p,ɬ0րE&sAuEBfB$^\3VFD5HĽx!,dWipVY
C5p"0O㿈D
pOZ
sjxD vÈ5x#, =6_hL ŧ"[a"t_\a~
1Cxz
&BLC0]̀\в\BAl``G=LLJ% H3X7_ҋDd
`b
c$A??3"`?2v W$]Á]n&+`!v W$]Á]n@X
wxcdd``> @c112BYL%bpuc0R`fɤWE.2ԱD+á6{ dkpenR~CP4v[m/<1@k*5nނ0opV
Pam
ĹH07!+40ݜRaXyib(pbdbR
,.I@X`?UvDd
tpb
c$A??3"`?2[ד7g؍<++`![ד7g؍<~ /
Ibxڝ]HSa=:ϖ؇˯L"
qm^lb[W0!VPyXHtY7!%EЍ$H!l9OY9à
n<JّĘ6bR&!maȮCs.ᣆ'TAF\ȣv〮Xw">fҍ͔8߮XhtOrx/?3cosGO1c63l8lr+3ruEVw\]Q{!6}CXP:ި@ƺ+G(B*鹫}إ l\$SmB&w$)Qd8i(cԎΜ֑:eHD&,!MIaţg2sBJHϼEPzY3PtJo*
̣Z*~=ǒ̗={ !Y+[Gk$ 6rMzapȐsuFapʐ{
8 Zlsܛ&I5=PLOY!욳tguADd
b
c$A??3"`?2yaK+g0+`!_yaK+@,@DxtxڕMhALtmVuЂRxھ),A
Pou*p`]
JSVuh_p}:p*ZGMsfna%Y Ma<;=O hh={M*. rY\IL.Rvtma%2ji9XhԮaEۜ FS*S]U9*ƶ2%ڔdUjP
u9y~9AJm$&`id)\^MCkhTZohöouոr$n_m~K} ~垇$nsnEšv9gY%f\$ѨfɬLqRN/BV)8!!VSl_!ʃrVFlf0=Dd
0
#A2o)n{g@).2!N+`!o)n{g@).2
{tls2#![xڵ1hSQϻB*M+PC!Y]l$HCUfcds}7&F8=nn32@
?
8QJAQX4&*VF!) ,
_Z;
}Qnb/?ŧx/=~p!!fw{ΜZa)LǴPcb&v<GU9saGba6Op߀\t]51_іf s &j7WTD7%8d]gVG(i(F)`_KB>Sr+B*^u)7ip+u[2s?v]Ǹ5ka];TeUjݾM$>$ebY$4x/v8f1EL¼b:gԏ!^()}QDna8p>슔Wl'lEK*"[.f=.](M(4W?[;<~{N6eV!!0"*^ K[3_F}/?vS=8aSjE{E쵯P`YL`$2ru{mG]\j9Ĳ
E쵥v)7:R:>z@Iq5Y8<%Ú!g
bj76"=8٪d~=;\5tjwGQ7GUqx]C=یn;o;[P9UtvܙʳykprLlT)܆cThRn8l!ںK[ j{ȆWgq5.F͊(b i{&Ƅ0rޣvBލP#28aTQ#_0f,VQUFYW$NQ[BA+BB]MLbBG^.y#zcUBs; }u<^[jO̕1Jyc?UͣϡnVu6j3nXu!桿ޠ37/}٤Dd
X\
1a0
#A2 SN MlU+`!SN Mlb7ZH9#!xڵkQ_潙5A*b BzETT*=DЋ/(JZJh$%գS3v~3폌WJ&HS3:ӧ9(5`$KTZV #Y&ɨphժO#<#uGYTotNu^HX~×Xq'Ky6*rEjUՏ<6GQj@uoǺнv@HT$0pPyAI'˿&GaM:ANj7nqU_i簨gPtsX:ZM$$If!vh55`55555#v#v`#v#v#v#v:Vl
t0&655`5555apFkd[$$Ifl֞V!T$`
t0&644
lapF$$If!vh55`55555#v#v`#v#v#v#v:Vl
t0&6,55`5555a$$If!vh55`55555#v#v`#v#v#v#v:Vl
t0&6,55`5555a$$If!vh55`55555#v#v`#v#v#v#v:Vl
t0&655`5555a>$$If!vh55`55555#v#v`#v#v#v#v:Vl
t0&655`5555apFkdJa$$Ifl֞V!T$`
t0&644
lapF$$If!vh55`55555#v#v`#v#v#v#v:Vl
t0&6,55`5555a$$If!vh55`55555#v#v`#v#v#v#v:Vl
t0&6,55`5555a$$If!vh55`55555#v#v`#v#v#v#v:Vl
t0&655`5555a$$If!vh555#v:Vl
t065$$If!vh555#v:Vl
t065$$If!vh555#v:Vl
t065$$If!vh555#v:Vl
t065$$If!vh555#v:Vl
t065$$If!vh55#v#v:Vl
t0655$$If!vh555#v:Vl
t065$$If!vh555#v:Vl
t065$$If!vh555#v:Vl
t065$$If!vh555#v:Vl
t065k$$If!vh55 #v#v :Vl
t655 /$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/k$$If!vh55 #v#v :Vl
t655 /k$$If!vh55 #v#v :Vl
t655 /$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/k$$If!vh55 #v#v :Vl
t655 /k$$If!vh55 #v#v :Vl
t655 /$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/k$$If!vh55 #v#v :Vl
t655 /k$$If!vh55 #v#v :Vl
t655 /$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/k$$If!vh55 #v#v :Vl
t655 /k$$If!vh55 #v#v :Vl
t655 /$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/k$$If!vh55 #v#v :Vl
t655 /k$$If!vh55 #v#v :Vl
t655 /$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/k$$If!vh55 #v#v :Vl
t655 /k$$If!vh55 #v#v :Vl
t655 /$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/$$If!vh555#v#v#v:Vl
t6555/SummaryInformation(q 8DocumentSummaryInformation8CompObjj4rlMicrosoft Word 10.0@@UO.@jiq@qYy՜.+,0lhp
KHometIŎ
The Youth Risk Behavior Survey (YRBS) is sponsored by the CDC and is designed to monitor behaviors that are related to causes Title
FMicrosoft Word Document
MSWordDocWord.Document.89q@@@NormalCJ_HaJmH sH tH DA@DDefault Paragraph FontRi@RTable Normal4
l4a(k(No List>@>S
Footnote TextCJaJ@&@@SFootnote ReferenceH*HHxBalloon TextCJOJQJ^JaJj@#j.
Table Grid7:V04 @24Footer
!.)@A.Page Number4@R4%Header
!1r59I\`]_`
LM[jpq<=cdzfg!!'!+!,!L!M!!!O(P(((((****rsX.Y.t.!1"1337799999 :<<>>1?6?9?F@FAFFFGFJFKFPFQFRFXF\F]FcFdFgFhFnFoFsFtFzF{FFFFFFyJzJNN9S:SeUfU]X^Xb[c[[[[[^^_(_)_=b>beehhhhhhhhhiiiii%i,i/i2i3i`ibieilisiviyizi{ii}i~iiiiiiiiiiiiiiiiiiiii)j+j.j4j;j?jBjCjDjEjFjGjHjIjJjKjjjjjjjjjjjjjjjjjkkmmKmVmemmmmmmmmmmmmmmmmmmmmmmnnnnn#n(n)n*n7nnKnPnQnRn]n^noo v
vyyƀǀ>?+,ghJWY^
%&')678:GHIKXYZ\ijkmz{~
'()+4568ABCENOPRcdefgl()*,5679BCDFOPQS\]^`ijkm~ی܌ߌ
klmo3457[\]_؎َێ
km*nƒǒ
0000000000000000x00x00000000 0p0p0p0p0p0p0p0p0p0p00 00 0p0 0p0p0p0p0p0p000p0p0(0p0p0p0p0p00p00 0p0 0000p00@0p000p0000 08080808080808080 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0$ 000@0p0@0x0@0@0 0 0 0 0$ 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0 0 0 0 0 0$ 0 0 0 0 0 0 0 0 0 0 0 0 0 0$ 000x000x0x0@00000x0x0x0x0X0000000000000000x0h0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 000x00x0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0000h00h0h000p00x0000000000000000000000000000000000 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0000@0x@0p@0@0@0p@0@00 @0@0@0@0@0@0@00$0000000000000000000000
0M!!!O((1?6?9?F@FAFFFGFJFKFPFQFRFXF\F]FcFdFgFhFnFoFsFtFzF{FFFFFNNfU]X^Xb[c[[[[^^_(_ehh/i2i3iiiiijjjjjjmmnnnnn#n(n)n*n7nnKnPnQnRn]n^nooJ()*,56BCOP\]ij~;00L;00;00Oy00Oy00@0@0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 ^>00@0@0@0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0 @0@ @0 O900O900O900O900O900O900 sO900O900O900O900O900{00O90 0O90 0O90 0O90 0{00O90
0 h{00{00L{00LOy00 @0 {00L{00LOy00 @0 {00LOy00Oy00 {00LOy00 @0 Oy0(0:Oy00 Oy0*0 Oy0*0 Oy0(0 Oy00Oy0*0 Oy00Oy00@ p0 Oy00Oy00Oy00p0 Oy00Oy00Oy00@0 O90:0O90@0O90@0
@0Oy00Oy004Oy00Oy00{00@0O90K0 lO90K0O90K0@0Oy004
@0Oy00Oy00Oy00
@0 Oy00bIH<
M9 Oy00<@0 Oy00tA0 Oy00@#0D Oy00WZ7:7 0}$/1114%'(6+G2389b:"@VDFKQ[[bgNimkr,uGwnze/+[NQRTUVWXYZ\]^_`bcdfvx~q+9RFKKKKKKKKKKLL6L7L\L]LR`RgRRVnk[uuuuuuuu6vQvvvvvw$wbwyyz!z6zRzozzzze=ߕ<dĖՖ@hu?Ę٘ܘ6^fmpƙ2AWdEOS[aeghijklmnopqrstuwyz{}Pqg{} !#!%!,!@!B!P(o(q(r((((((***+++,,,sc[[[[[[^^^^__::::::::::::::::::::::!$+4!!l,"$nz]*[k"}_@0(
B
S ?8(D")Dd*D$+D
,DD.D4l /D40D\1Dt2Do3D4D 5D"6D~7Dta8DԾ9D:D;D\dD4
?D@D,ADBDlCD4nDD
ED
FD<
GDRHDRID$SJDjKDjLDkMD\kNDTODPDQD.RD<SDTDUDVDWD<XDYDZD
[D
\D4
]Dt
^D
_D
jjxII QQt/t/HiHiLijjj}}"mm˄ŅŅԅ))66>
"!#%$&(')*+,./01234657tOO&&WW//JiNiNijjj3}3} $$z~~Ʉ̈́̈́҅օօ ##44=HH
"!#%$&(')*+,./01234657=6*urn:schemasmicrosoftcom:office:smarttags PlaceName=7*urn:schemasmicrosoftcom:office:smarttags PlaceTypeB**urn:schemasmicrosoftcom:office:smarttagscountryregion84*urn:schemasmicrosoftcom:office:smarttagsCity98*urn:schemasmicrosoftcom:office:smarttagsplace9'*urn:schemasmicrosoftcom:office:smarttagsState87684848484848*84'84'8484'84'848'84'84'84'8484'84848*867 + . G J W Z \ _ S
V
W
Z
^
a
b
e
/6>D
[dop##**++++++_a::M=U=pLzLNNQQeQTUVV>X@X8^:^ccccllnnuuxxxx9}A}ǀ̀,5Ɔhnǎǒ\d '!)!!!$$(((((((*7***P.S.>?XE`EeJhJPPfUU>XAX[\,\0\8^;^;`A`bbeehhjjkkm
m?TU˂BIElOą+h܇*%(8Dǒ3333333333333333333333333333333333333333333333333M[jqg~ !&!'!,!C!M!Q***++, ,s....=6=699999 :1?@CCXEaEEFFFNN[[^*_hii3i`iiiiiiiiiiiiiiii)jKjjjjjjjjjjjjjjjjKm^n v
vǀ?,mP~4h݇^l*ߌmo7]_َ
ǒޒǒ
W0=g0$[Rh%C2iTUKQ/ (
8
_5W6
mG
Z
Qc
&$/#lcVaa2 = V}!b")K#N$$%n&M&4,'5'4(Y~)h3*#+4+,$7,S,X,.Z.!/r/:x0{01627q2)s689C9qy9i(;V;<+=QV=&2>%d>SASAXDPCEW+GmHKOKd NNSDTqTUwT`
U
NWXMY VYq}Z*[BD[N]b]&_^._H_`1`ad5dhud_fVgXjSjkm$m^Amnz{_>Te~g(irQ0i5f%6xHnpLU^NKZLatn^60PRor,lahl/ =CS]S/@\Q{TbqY,#34HmxVQtM2
0
8@Re"`Z P~r,y"m_xzT][FDKX.u\fDf#L:v=_E=W3uS2he=a>FX"dMt \0~,r'ZZ1?6?9?F@FAFFFGFJFKFPFQFRFXF\F]FcFdFgFhFnFoFsFtFzF{FFhhhhhhhiiiii%i,i/i2i3i`ibieilisiviyizi{ii}i~iiiiiiiiiiiiiiiiiiiii)j+j.j4j;j?jBjCjDjEjFjGjHjIjJjKjjjjjjjjjjjjjjjjjKmVmemmmmmmmmmmmmmmmmmmmmmmnnnnn#n(n)n*n7nnKnPnQnY^
%&')678:GHIKXYZ\ijkmz{~
'()+4568ABCENOPRcdefgl()*,5679BCDFOPQS\]^`ijkm~ی܌ߌ
klmo3457[\]_؎َێ
@jjPqCjjT$LW W*+l`@``,@`6`p@`p@UnknownGz Times New Roman5Symbol3&z Arial5&zaTahoma"qh턆YyIYyIY24dŎŎ3QH)?2 ~The Youth Risk Behavior Survey (YRBS) is sponsored by the CDC and is designed to monitor behaviors that are related to causes Preferred CustomerCarl Setzer