• No results found

Table 5.15: Posterior class probabilities and observed frequencies of drug use in the sample

Class n. xt

1

xt

2

xt

3

xt

4

xt

5

C1 low use 1412 .01 .17 .38 1.96 2.77

C2 desisters 46 11.33 20.14 33.02 28.5 16.90 C3 adolescence lim. 93 .2 15.55 60.86 48.94 52.96

Sample x 1551 0.36 1.52 4.51 4.94 5.79

Comparing Figure 5.9 with the table above it is possible to approximately check the validity of the estimated trajectories, with the limitations discussed above. The frequencies for the low-use class follows a slow increasing pattern that remains at low levels of consumption all across the measured time span. This is confirmed by the trajectory of the class and also supports the already mentioned idea that the members of this group are not only non users, but also experimental consumers; this is made clear by the increasing frequencies which reach a mean value of nearly three times in the last 12 month in 2006.

The mean frequencies for the class of high-level desisters do not perfectly match the estimated curve as in the first case. Here, in fact, a desisting process is only visible from the third time point onwards. A better picture is instead given by the adolescence-limited

group. Their measured frequencies approximately match a bell-shaped curve as depicted in Figure 5.9. Here also, from a mean value of nearly zero (0.2 times) they reach a peak of 60 times in the last twelve months in 2004, and then slowly desist to a level at about 50 times in the last two measured years. More in general, it can be stated that the observed frequencies match the estimated classes very well for the larger group of low users and for the adolescence limited, and slightly worse for the desister class. Keeping in mind the possible pitfalls involved in such a transformation, the interpretation of the observed frequencies suggests that the smaller class of desisters could be a statistical representation of a more heterogeneous group of subjects characterized by unusual developments and high level of consumption.

5.4 Conclusion

The number of subjects using illicit drugs is small compared to the large sample size.

Nevertheless, some of those using such substances also reported high frequency of use (see Chapter 4). Furthermore, the aggregate values suggest that this behavior does not remain constant across the covered time span, but rather increases at the beginning and tends to recede at the last measurement point. This fluctuation in the aggregate level of the frequency and prevalence of drug use was a good premise for further investigations using longitudinal analysis techniques such as GMM. In a first step LGM analysis was car-ried out, which suggested that a curvilinear development better represented the aggregate development of drug use behaviors in the sample. The found trajectory showed an increas-ing pattern across the first three time points which then stabilized at the remainincreas-ing two measurement points. Furthermore, variability was found in all three estimated random effects, which suggested that this variability could be further investigated by means of GMM. According to the model-fit indices and substantive interpretation of the graphical results, three classes were deemed necessary to represent the sample development in drug consumption. In fact, compared to the two-class model, the newly estimated third group isolated a specific pattern of development otherwise hidden within the low-user class. A further fourth class, which could only be estimated by a LCGA model, was clearly a consequence of the restricted zero variance on the random terms, and thus superfluous.

The results of the GMM showed high stability across different model specification, espe-cially those concerning the variance of the random terms of each single trajectory. For this reason the model with variance for the random intercept was preferred because of its parsimony and a slightly better reproduction of the observed values. According to this results, the development of drug use in the sample can be represented by a large low-user class (91%), a class of adolescence-limited consumers (6%), and a smaller desister class (3%).

The low-user class is characterized by being the more representative group in the sampled population, and thus arguably the most common marijuana use developmental pattern during adolescence among youths in Duisburg. Although a considerable amount of sub-jects in this group has never used drugs in the analyzed five years period, some of them have reported experimental and sporadic use. The average amount of times that members of this group used drugs in the last 12 months never goes over the three times a year, al-though the estimated trajectory shows a constant increase across the measured time span.

This result is also supported by recent studies on trajectories of drug use. Although some studies report separate trajectories for low users and abstainers (Ellickson et al., 2004;

Schulenberg et al., 2005; Tucker et al., 2005), many other also included abstainers and experimental users into a single class (Brown et al., 2004; Jackson et al., 2008; Brook et al., 2011). The results in this work support the idea that marijuana use is a common experience among the general school population, a phenomenon which tends to increase over time in late adolescence and that involves a large proportion of the youth population.

5.4. Conclusion

Similarly, also the amount of abstainers remains large, even if hidden within the variance of the low-user trajectory.

The adolescence-limited class is the second largest group in the sample and is represented by a well-known bell-shaped pattern of development. For these youths, drug use behavior is clearly limited to the adolescence period, although by the end of the measured time span they do not yet completely desist from use. Their pattern of development clearly begins at the age of 13, increases rapidly in the next two years to an approximate average level of 20 times a year, and decreases at a slower pace thereafter. In terms of frequencies they reach the highest point at age 15, and although they clearly diminish the average use, by the age of 17 they still remain the group with the highest frequency of use. If compared to recent literature on the subjects, Flory et al. (2004) and J. Guo et al. (2002) also found a similar trajectory with a similar age-development. However, compared to other studies (Jackson et al., 2008; Brook et al., 2011), the adolescence-limited group found here seems to desist earlier than expected. In any case, the adolescence-limited class supports the reassuring perspective that high frequencies of marijuana, and drug use in general, in the sample are a temporary phenomenon, limited to a specific and short phase in life.

The last group is also the smallest in the sample and includes those subjects who reported a decreasing use of drug over the five year period. Their trajectory clearly differentiate itself from the other two for an earlier start and a consequent slow decreasing trend. At the first time point, at the age of 13, they already report fairly high frequencies of drug use (approximately an average of 10 time a year), which, although they tend to decrease in the following years, by the age of 17 they still maintain a level which can be defined as “occasional consumers”. A similar pattern has also been found in many of the studies cited in Table 2.1. Ellickson et al. (2004) and Tucker et al. (2005), for instance, also found a group of subjects who reported high frequencies of marijuana use prior to the age of 13 and decreased thereafter, still remaining occasional users by the age of 23. Also Windle and Wiesner (2004), analyzing a shorter period of time, found a small group of adolescents with already high frequencies of marijuana use at the age of 15, and a de-creasing pattern thereafter. However, in the remaining studies, a dede-creasing trajectory was associated with an adolescence limited development of marijuana use. In this study, by the age of 17 they are still involved in marijuana consumption and it cannot yet be stated, whether this pattern is limited to adolescence only. In any case, this group retains its importance, especially because it enables the identification of an early-starter group that by the time of late adolescence is still involved in drug consumption.

It can be concluded that the resulting trajectories have found empirical and substan-tive support also in the principal current studies. The general trend that can be identified is one of a general reduction in the frequencies of use with the approach of late adoles-cence. Although the low-user group is still characterized by an increasing pattern, those who during adolescence were characterized by elevated marijuana consumption are char-acterized by a clear declining trend. Whether or not this trend will continue over the next years, can only be answered by means of the introduction in the analysis of successive panel waves.

Chapter 6