China
Africa
Explore chapters and articles related to this topic
Weighting and Complex Sampling Design Adjustments in Longitudinal Studies
Published in Jason T. Newsom, Richard N. Jones, Scott M. Hofer, Longitudinal Data Analysis, 2013
Shayna D. Cunningham, Nathalie Huguet
All the replication-based methods require the creation of a set of replication weights to ensure that each replicate appropriately represents the same population as the full sample (Carlson, 1998). There are various ways to create replicate weights. For example, for the balanced repeated replication method, a simple procedure is to double the sample weights of observations in the selected PSUs for a replicate to make up for the half of PSUs not selected. A variant procedure is to use 2 – k with k (0 ≤ k < 1) times of the sample weights (Judkins, 1990). For the jackknife repeated replication method, the replicate weight is obtained by dividing the sum of the weights in the retained PSUs by a factor of (1 − wd/wt), where wd is the sum of weights in the deleted PSU and wt is the sum of weights of all PSUs in that stratum. This procedure is included in software that uses this approach, so there is no need to generate the weight as in the balanced repeated replication methods. For the bootstrap repeated replication method, various procedures have been suggested (Sitter, 1992). One way to handle the replicate weight is to create the replicate by selecting the bootstrap sample with the selection probability proportional to the magnitude of the sample weight. Depending on the statistical software used, the replication weights will need to be entered as data or will be generated by the program from the specifications provided. To create the replication weights or to specify the procedure, the analyst must understand the sample design features. Fortunately, most surveys requiring the use of a replication method provide the replicated weights to users.
A longitudinal assessment of change in marijuana use with other substance use problems
Published in The American Journal of Drug and Alcohol Abuse, 2018
Namkee G. Choi, Diana M. DiNitto, C. Nathan Marti
To account for the PATH Study’s multi-stage stratified area probability sampling design and nonresponse adjustments, Stata/MP 15’s svyset function was used to incorporate W2 weights and variance estimation methods (balanced repeated replication-Fay’s method as specified in the PATH Study) in all analyses. For all subsample analyses (e.g. W2 marijuana users), Stata’s subpop command was used to ensure that variance estimates incorporated the full sampling design. For descriptive analysis of sample characteristics (by marijuana nonuse/use and use frequency), we used χ2 (omnibus and paired for multigroup comparisons), one-way ANOVA, and t-tests. We then used multinomial logistic regression analysis to test H1 (i.e. determine whether W1-W2 never users and W2 quitters differ from W1-W2 ex-users) and H2 (i.e. determine whether W2 new, resumed, and continued users differ from W2 quitters), and binary logistic regression analysis to test H3 (i.e. determine whether frequent users differ from less frequent users). Variance inflation factor diagnostics, using a cut-off of 2.50 (35), indicated that multicollinearity among covariates was not a concern. Multinomial logistic regression results are reported as relative risk ratios (RRR) with 95% confidence intervals (CI) and binary logistic regression results are reported as adjusted odds ratios (AOR) with 95% CI. Significance level was set at p < .05.
Comparing American college and noncollege young adults on e-cigarette use patterns including polysubstance use and reasons for using e-cigarettes
Published in Journal of American College Health, 2020
Anne Buu, Yi-Han Hu, Su-Wei Wong, Hsien-Chang Lin
Independent sample t-tests and Chi-square tests were conducted to examine group differences in continuous and categorical variables, respectively. We compared the college and noncollege samples on sociodemographic and substance use information. Among current e-cigarette users in both samples, we compared their patterns of cigarette use, e-cigarette use, and polysubstance use, as well as their reasons for using e-cigarettes, smoking rules at home, and perception of harmfulness. In order to obtain nationally generalizable results, all parameters were estimated with PATH Wave 1 survey sample weights and the balanced repeated replication method (Fay = 0.3) was used to compute the standard errors.
Racial/ethnic group comparisons of quit ratios and prevalences of cessation-related factors among adults who smoke with a quit attempt
Published in The American Journal of Drug and Alcohol Abuse, 2022
Dana Mowls Carroll, Ashley Cole
Analyses were conducted using SAS version 9.4, with appropriate survey procedures and sampling weights. Variances were estimated by the balanced repeated replication method (34), with Fay’s adjustment set to 0.3 to increase estimate stability (35), as this approach is recommended by the PATH Study Userguide (29).