Research Questions |
Analysis Variables |
Statistical Model |
(1) Identify and examine predictors and factors associated with dichotomized outcome measures (e.g., early onset of drug use, early sexual initiation, mortality). (2) Evaluate a treatment effect in an experimental study (e.g., an experimental device, a treatment, an intervention program). |
The dependent variable is a dichotomized variable (e.g., early age of first drug use, age<=15 vs. age>15). Explanatory covariates are time-invariant |
Logistic regression models |
(1) Study the temporal patterns of an event, accommodating varying lengths of follow-up time among subjects (2) Determine the time to the occurrence of an event (e.g. onset of drug used, sexual initiation, abstinence, relapse after treatment). (3) Estimate the impact of risk and protective factors on the time to occurrence of an event. (4) Determine the effect of time-varying explanatory variables. |
The dependent variable is time to the incident of the event studied. Explanatory covariates can include: (a) time-invariant covariates (e.g., gender, race, education, marital status, employment status, primary drug). (b) time-varying covariates. (e.g., level of drug use across years, incarceration across age) |
Time to event (survival analysis) |
(1) Explore temporal and dynamic changes on an outcome measure across time (e.g., level of drug use over time, sexual risk over time). (2) Estimate the shape of an outcome trajectory (e.g., intercept and slope of the outcome over time). (3) Identify and examine the impact of risk and protective factors on the shape of the outcome trajectory. |
The dependent variables are multiple measures of a variable across time (e.g., days of drug use across years). The dependent variable can be continuous, count or dichotomized /categorical. Explanatory covariates can include (a) time-invariant covariates (e.g., gender, race, education, marital status, employment status, primary drug); (b) time-varying covariates. (e.g., level of treatment services across years, duration of incarceration across age, degree of criminal involvement across time) The model estimates the shape of trajectory across time. Two levels of data, where Level 1 consists of repeated observations within subjects and level 2 captures the influence of fixed individual-level factors, can be analyzed as mixed or multi-level models. Or as repeated measures regressions (general or generalized linear models). Determinations: time invariant, time-varying, continuous, interval or ration-level measurement. |
Growth curve model; random effects regression; mixed effects regression; generalized linear or non-linear models (including GEE). |
(1) Identify distinctive trajectory patterns of an outcome for subgroups of the study sample. (2) Examine and compare subjects’ characteristics among the identified trajectory groups (e.g., demographics, drug use history, mental health status, sexual risk behaviors). |
The dependent variables are multiple measures of a variable across time (e.g., sexual risk across age). The dependent variable can be a continuous, count, or dichotomized/categorical measure. Explanatory covariates can include (a) time-invariant covariates (e.g., gender, race, education, marital status, employment status, primary drug); (b) time-varying covariates. (e.g., level of treatment services across years, duration of incarceration across age, degree of criminal involvement across time). The model estimates the shape of trajectory across time and identified latent classes (trajectory groups). Subjects within the same classes have a homogeneous trajectory pattern. Subjects among different classes have heterogeneous trajectories. |
Growth mixture model or group-based trajectory model |
Study Design |
Timing and frequency of measures |
Research Aims |
Analysis Variables |
Statistical Model |
Cross-sectional |
One time assessment All measures collected at the same time point. |
(1) Identify risk and protective factors associated with a continuous outcome measure (e.g., ASI score, treatment retention, sexual risk score) (2) Examine or compare group differences (e.g., treatment modality, ethnicity) in the outcome measure. |
The dependent variable is an interval or normal variable (e.g., ASI score, treatment retention). |
Multiple regression models |
(1) Identify risk and protective factors associated with a dichotomous/categorical outcome measures (e.g., early onset of drug use, early sexual initiation). (2) Examine or compare group differences (e.g., treatment modality, ethnicity) in the outcome measure. |
The dependent variable is a dichotomous variable (e.g., early age of first drug use, age<=15 vs. age>15) or multi-category variable. Explanatory covariates are time-invariant. Measures of the outcome and covariates were collected at the same time point. |
Logistic regression models |
Study Design |
Timing and frequency of measures |
Research Aims |
Analysis Variables |
Statistical Model |
Longitudinal study design with one follow-up (e.g., pre-test/post-test) |
Two assessments over time. Measures were collected at intake and follow-up. The main outcome measures were collected at Follow-up. The explanatory covariates (predictors) were collected at intake or at follow-up but representing period between intake and follow-up. |
(1) To identify factors that may predict an interval-level or normal outcome measure (e.g., ASI score, number of arrests). (2) To evaluate a treatment effect on the outcome (e.g., The impact of an experimental device, a treatment, or an intervention program on the ASI score). |
The dependent variable is a continuous variable (e.g., ASI score, treatment retention) collected at the follow-up. Explanatory covariates are time-invariant, typically from intake or the period between intake and follow-up. |
Multiple regression models |
(1) To identify factors that may predict a dichotomized/categorical outcome measure (e.g., relapse, crime involvement, mortality). (2) To evaluate a treatment effect on the outcome (e.g., impact of an experimental device, a treatment, and an intervention program on status of successful recovery). |
The dependent variable is a dichotomized/multi-category variable (e.g., status of successful recovery, yes vs. no). The dependent variable was collected at the follow-up. Explanatory covariates are time-invariant. |
Logistic regression models |
Study Design |
Timing and frequency of measures |
Research Aims |
Analysis Variables |
Statistical Model |
Experimental study design with two assessments |
Pre and post intervention (treatment) assessments |
Evaluate the intervention effect on the outcome. |
The outcome variable is an interval or normal measure (e.g., ASI score) The outcome variable is categorical measure (e.g., positive vs. negative urine test) |
t-test ANOVA Chi-square test; multinomial models |
Evaluate the intervention effect on the outcome controlling for some confounding factors (e.g., gender, ethnicity, severity at intake) |
The outcome variable is an interval or normal measure (e.g., ASI score) The outcome variable is categorical measure (e.g., positive vs. negative urine test) |
ANOVA; ANCOVA; Mixed models Logistic regression models; GEE |
||
Experimental study design with multi-level data |
Study design involved with a multi-level sampling scheme (e.g., individuals clustered within programs, such as patients are nested with a treatment program; Children are nested with Mothers) |
Evaluate the main intervention effect with incorporation of level-1 (individual-level) and level-2 (program-level) of factors simultaneously. |
The outcome variable is an continuous or binary/categorical measure (e.g., ASI score; abstinence vs. drug use). Level 1 consists of individual-level of measures, and Level 2 captures the influence of program-level factors. |
Hierarchical (multilevel) linear models |
Study Design |
Timing and frequency of measures |
Research Aims |
Analysis Variables |
Statistical Model |
Experimental study design with multiple assessments across time |
Multiple assessments across time |
(1) Study the temporal patterns of an event, accommodating varying lengths of follow-up time among subjects (2) Determine the time to the occurrence of an event (e.g. onset of drug used; sexual initiation, abstinence, relapse after treatment). (3) Evaluate an intervention effect (comparing survival curve by intervention group) (4) Estimate the impact of risk and protective factors on the time to occurrence of an event. (5) Determine the effect of time-varying explanatory variables. |
The dependent variable is time to the occurrence of the event studied. Explanatory covariates can include (a) time-invariant covariates (e.g., gender, race, education, marital status, employment status, primary drug); (b) time-varying covariates. (e.g., level of drug use across years, incarceration across age) |
Survival analysis |
Experimental study design with multiple assessments across time |
Multiple assessments across time |
(1) Explore temporal and dynamic changes in an outcome measure across time (e.g., drug severity over time). (2) Evaluate an intervention effect (comparing the outcome trajectories by intervention status) (3) Identify and examine the impact of risk and protective factors on the shape of outcome trajectory. |
The dependent variables are multiple measures of a continuous or binary/categorical variable across time (e.g., days of drug use across years). Intervention status should be the main covariate. Additionally, explanatory covariates can include (a) time-invariant covariates (e.g., gender, race, education, marital status, employment status, primary drug); (b) time-varying covariates. (e.g., level of treatment services across years, duration of incarceration across ages, degree of criminal involvement across time). The model estimates the shape of trajectory across time. |
Growth curve model |
Study Design |
Timing and frequency of measures |
Research Aims |
Analysis Variables |
Statistical Model |
Retrospective (Longitudinal) study design OR Prospective (Longitudinal) study design |
Two or more assessment points and retrospectively collected longitudinal data (e.g., Natural History data) OR Multiple assessments across a long-term time period. Longitudinal data were collected prospectively |
(1) Explore dynamic changes on an outcome measure across time (e.g., level of drug use over time, sexual risk over time). (2) Depict and estimate shape (e.g., intercept and slope) of trajectory of the outcome. (3) To identify and examine the impact of risk and protective factors on the shape of trajectory. |
The dependent variables are multiple measures of a continuous variable across time (e.g., days of drug use across years). The model estimates the shape of trajectory across time. Explanatory covariates can include time-invariant covariates (e.g., gender, race, education, marital status, employment status, primary drug). Multiple measures across time (change with time) can be included as time-varying covariates. (e.g., level of treatment services across years, duration of incarceration across ages, degree of criminal involvement across time) |
Growth curve model |
(1) Identify distinctive trajectory patterns of an outcome among the study sample. (2) Examine and compare subjects’ characteristics among the identified trajectory groups (e.g., demographics, drug use history, mental health status, sexual risk behaviors). (3) The prospective longitudinal study will allow one to identify and characterize homogeneous groups, and risk factors that are longitudinal related to trajectory |
The dependent variables are multiple measures of a continuous variable across time (e.g., sexual risk across ages). The dependent variable can be a continuous, count, or dichotomized measure. The model estimates the shape of trajectory across time and identified latent classes (trajectory groups). Subjects within the same classes have a homogeneous trajectory pattern. Subjects among different classes have heterogeneous trajectories. Explanatory covariates can include time-invariant covariates (e.g., gender, race, education, marital status, employment status, primary drug). Repeated measures (those change with time) can be included as time-varying covariates. (e.g., level of treatment services across years, duration of incarceration across ages, degree of criminal involvement across time). |
Growth mixture model or group-based trajectory model |
||