 This is the first presentation of a series of three where we'll go into more details about the goals and aims of latent transition analysis and how to conduct it. First, I will provide an example of a research problem to illustrate what latent transition analysis can do. I will then illustrate the purposes of latent transition analysis. And I will emphasize how latent transition analysis is basically a person-centered approach to investigate change across time. Latent transition analysis allows to specify measurement models at each time point, which represent the different types of groups or participants in a study. And then it allows to investigate how individuals move from one group to another across time, that is structural changes across time. So I'll start with this example. Let's assume we ask adolescents to report if they have frequently used any of these substances in the last month, alcohol, cannabis and so on. Adolescents have 14 years of age, the first time we ask them these questions. And just for the sake of the example, let's assume there are two genders, males and females. We observe different patterns of substance use in this sample. And we may be interested in classifying these individuals into fewer groups that represent key differences in patterns of use that we observe. Or as those groups may represent different propensities to substance use of the sample. And for this purpose, we might advise some arbitrary rules. For example here, I grouped in red the participants that reported use of two different substances. And I grouped in orange those that reported use of one substance only, whereas in black I grouped those that did not use any of these substances. But these groups are made up based on arbitrary criteria I made up. So imagine that we ask the same questions a year later to the same adolescents. At age 15, we may notice even more patterns of behaviors. And for example, the patterns of behaviors I circled here were not observed at age 14. And once again, we may be interested in classifying individuals into fewer groups that can adequately represent the multiple patterns we observe in the sample. And I may use the same arbitrary rules I made up for the sample at age 14, or I may create additional rules. For example, the bright red individuals here are those that report the use of multiple substances. However, arbitrary rules are inadequate because they may not represent key differences across individuals and they are not reliable and meaningful across different studies. Furthermore, when we are interested in classifying individuals into different groups based on the patterns of substance use, we're also often interested in investigating whether individuals change or not. And if they change, what are the patterns of change? For example, based on the gateway theory, I might expect that I might be interested in checking and investigating whether individuals that use cannabis at 14 years of age move into categories of use that involve other substances as well. And again, if the categories we identified are based on arbitrary rules, we are often not able to provide reliable descriptions of change across time. The latent transition analysis is a statistical method that, based on probability rules, allows to answer these questions. In other words, latent transition analysis allows to identify fewer groups that represent variability of behavior patterns at each age and investigates the patterns of change across time. What are the probabilities of individuals remaining in the same categories or moving into others across time? And also allows to investigate what are the factors, for example gender, that can affect change over time? So here I will describe in more details what are the purposes of latent transition analysis. Latent transition analysis extends the latent class model to repeated measures and longitudinal data. Latent class analysis is a person-centered approach. This means that when we observe a sample that shows different patterns of behavior as in the example of adolescent substance use, we assume that what explains the interpersonal variability in the behaviors we observe is the fact that there are different categories of individuals who share the same propensity for displaying a pattern of behavior, for example the same propensity to use some drugs. At each time point we want to identify the groups or classes that can adequately explain the differences in behavior we observe across participants. And in this example I am assuming that at each age there are two underlying groups of adolescents, for example users and abstainers, people that don't use substances. So this is the measurement model I specify for the data and example. And the basic measurement model is assuming that there are different typologies, different types of individuals at each age and those differences between individuals explain the different patterns of behavior we observe. Once I have identified the categories of participants at each age, my question is how adolescents move from one category to another across time or if they move at all. For example, what is the probability that an abstainer at 14 years of age will become a substance user at 15 years? And what can affect this probability? For example, our maze more likely to move from being abstainers at 14 to becoming substance users at 15 years of age. Data transition analysis allows to answer all these questions using probability methods and therefore provides transparent and robust formal methods for answering questions about individual change over time. So I have said that data transition analysis firstly provides measurement models for data collected at repeated time points and I will illustrate now the main characteristics of the measurement models. So the data transition analysis extends data and class analysis to data collected over time or longitudinally. And I will summarize some of the key characteristics of data and classes here. If you want to know more about data and class analysis, I have prepared another resource for NCRM that you can use. So basically starting from the example, we observe variability in 14 year old responses to questions about substance use. Some adolescents report frequent use of alcohol and cannabis. Other report frequent use of alcohol or cannabis and most do not report any frequent use of these substances. The first goal of data transition analysis is to apply a latent class measurement model to each data collection point and therefore identify the number of underlying classes that can adequately explain the behavior patterns we observe. We cannot observe these classes directly, but we can inform them using probability rules and use them to assign participants to classes based on their behavior. Data and class analysis is a person centered method because it is focused on classifying persons and individuals. So let's assume that in this example three classes can optimally explain the patterns of behaviors we observed when adolescents are 14 years of age. These latent classes are supposed to be the underlying causal factors that explain the patterns of behaviors we observe. The patterns of substance use are explained by underlying typologies of individuals that differ in their propensities for substance use. For example, there may be a class of individuals I called users that share propensity for frequent use of different substances whereas individuals in the class I called experimenters display propensity to use some substance of choice. And those I have called abstainers tend to avoid frequent use of any substance. Now key assumptions of latent class analysis are that the classes we identify are exhaustive, which means that all the individuals in the sample belong to one of the latent classes. Each adolescent for example here will be either in the users or the experimenters or the abstainer class. The classes are also mutually exclusive, so an individual will belong to only one class, for example an abstainer cannot be an experimenter. So these classes are really types, typologies of individuals that share the same propensity to display a specific behavior pattern. However, latent class analysis is a probabilistic model, so the association between the latent classes and the behaviors we observe, which we call indicators, is observed with error. So for example individuals in the abstainers class may have 96% probability of not using alcohol, but there is still a 4% probability that they may use it. Consequently, individuals membership to the latent classes are also uncertain. We do not have the certainty that when we allocate an adolescent to the abstainer class, the adolescent actually belongs to that class, since the behaviors we observe are observed with errors. Latent class analysis allows to estimate that an individual may have say 89% probability of belonging to the abstainer class, but that means there is still 11% probability that the individual may belong to another class. And it is important to consider and control for this uncertainty when we regress latent class affiliation to predictors, as I will emphasize in other presentations. As we find the satisfactory measurement model for participants at one age, we should try to find another satisfactory one for the next time point, in this example, age 15. We might be tempted to assume that the same three classes we identified at age 14 could explain variability in adolescence behavior when age 15. However, we might also find that the patterns of behavior we observe at age 15 are more complex. To give an example, here there may be pattern emerging where adolescence also frequently use ecstasy in combination with alcohol and cannabis. And the analysis may tell us that three classes are no longer sufficient to explain the patterns of behavior observed, and we need four classes at 15 years where a new class of abusers emerge. This example emphasizes an important characteristic of latent transition analysis, that is, latent transition analysis allows to identify underlying categories that emerge at different time points. In other words, it allows to identify new behavior organizations that may emerge across time, as well as qualitative changes across development. This is an important characteristic that distinguishes latent transition analysis from other approaches to longitudinal data. For example, linear growth models that are more preoccupied with investigating changes in the level or degree of a behavior. Latent transition analysis is more concerned with changes in the organization of behavior, or it can be applied to questions about changes in the organization of behavior. But together with the measurement model, latent transition also investigates the structural relationship between the latent classes that explain observed behavioral patterns. It does so by considering the multinomial logistic regressions between latent classes at consecutive time points. For example, by regressing the latent classes at age 15 or those at age 14, we can investigate what are the associations between these patterns of behavior. In particular, if we do consistently identify the same underlying classes at different time points, for example, abstainers, we can investigate continuity. And that is what is the probability that abstainers at age 14 will remain abstainers at age 15. At the same time, the analysis can tell us about discontinuity. What is the probability that abstainers at age 14 will transition to a different class characterized by other patterns of substance use? For example, what is the probability that an abstainer 14 years will move to the abusers class at age 15? Furthermore, latent transition analysis allow us to identify to investigate the role of covariates. For example, is gender associated with latent classes affiliation at age 14? Do females vary in the probability of being abstainers compared to males? And do females and males vary in their membership at age 15 once we control for membership at age 14? And as I will illustrate in the third presentation, latent transition analysis can also provide answers to more complex questions. For example, questions about whether the transition probabilities vary by gender. Do females and males display different patterns of change across time? To summarize, the latent transition analysis is a person-centered approach applied to repeated measures and longitudinal data. There are two main goals of latent transition analysis. Latent transition analysis provides a person-centered measurement model that allows to identify subgroups of individuals that make up a sample at each measurement occasion. These subgroups are different types, classes of individuals that share the same propensity to display a pattern of behavior. And these propensities differ from those of individuals in other classes. Importantly, this person-centered approach allows to identify classes and behavior organizations that emerge over time. As well as a measurement model, latent transition analysis also provides ways to investigate structural relationships between the underlying latent categories at different time points. It can thus investigate continuity and discontinuity across development, that is how individuals may transition from one category of behavior to another one, and is therefore ideal for application to stable theories of development. Latent transition analysis fulfills these goals using probability methods that are robust and transparent. In the next two presentations, I will delve more into these methods. Thank you very much.