Creator: Staudt, Andreas; Baumann, Sophie
Contributor: Staudt, Andreas; Baumann, Sophie
Funding: German Research Foundation (BA 5858/2-1, BA 5858/2-3)
Title: Primary data on “Sensitivity analyses for data missing at random versus missing not at random using Latent Growth Modelling: A practical guide for randomised controlled trials”
Year of Publication: 2022
Citation: Staudt, A., & Baumann, S. (2022). Primärdaten zur Studie “Sensitivity analyses for data missing at random versus missing not at random using Latent Growth Modelling: A practical guide for randomised controlled trials” [Files auf CD-ROM]. Trier: Psychologisches Datenarchiv PsychData des Leibniz-Institut für Psychologie ZPID.
The data come from the PRINT study (“Testing a proactive expert system intervention to prevent and to quit at-risk alcohol use”), a randomized controlled trial. The sample of alcohol consumers from the general population (N = 1646) was randomized into anintervention and control group. All alcohol consumers were included in the study, regardless of the amount consumed. Standardized assessments took place at baseline, 3, 6, 12 and 36 months. The intervention group received three individualized feedback letters at baseline, after 3 and 6 months. The letters were automatically compiled by a computer-based expert system according to predefined decision rules and were based on the self-report data of the study participants at the respective measurement points. The letters contained individualized feedback on alcohol consumption, alcohol-related risk, motivation to change and other psychological variables (self-efficacy, decision balance, behavior change strategies). The intervention was based on the Transtheoretical Model of Behavior Change. The control group did not receive any feedback. The aim was to reduce the average amount of drinking after 12 or 36 months.
Research Questions/Hypotheses: The aim of this work was to illustrate sensitivity analyses for different missing data mechanisms (missing at random vs. missing not at random) using latent growth curve modelling. This can be used to answer the question: How do conclusions about intervention efficacy change if one alters the assumptions about the process that resulted in missing data?
Research Design: Fully standardised survey instrument (provides question formulation and answer options); repeated measurement
The information in the data set was completely self-reported, with the exception of the randomised group membership (“aprint”)and the variable “rgroup”, which contains each person’s membership in one of 16 possible patterns of missing values. Allquestions had a closed response format. Only age (“age”) was asked with a free text field. The question on “sex” had two responseoptions (male and female). To record the educational background, the participants were asked about their highest school-leavingqualification. For this purpose, they were presented with an exhaustive list of possible degrees, which were then transferred intothe dichotomous variable “edu”. The variable “together” (relationship status) was formed from several sub-questions, namelyquestions on marital status (single, married living together or separated, divorced, widowed), current relationship (if not married)and living situation (do you live with your partner?). The variables “aauditc” to “eauditc” each contain the AUDIT-C score, thesum score of the first three questions of the Alcohol Use Disorders Identification Test. The AUDIT-C score at baseline (“aauditc”)was used to form the alcohol-related risk “arisk”, using gender-specific cut-offs (4 or more for women and 5 or more for men).Further information can be found in the published study protocol (Baumann et al., BMC Public Health 2018).
Data Collection Method:
Survey in the presence of an investigator
- special apparatus or measuring instruments, namely
Baseline: Questionnaire on tablet PCs; All further measurement time points: Computer-assisted telephone interviews
Survey in the absence of an investigator
- Telephone survey
- online survey
Population: Alcohol consumers from the general population
Survey Time Period:
5 measurement points: T0 (baseline), T1 (3 months later), T2 (6 months later), T3 (12 months later), T4 (36 months later).
Sample: Full survey; others: Over a period of 2.5 months, all clients appearing in the waiting area of the registration office in Greifswald were proactively approached by study assistants. All clients who met the inclusion criteria and provided written informed consent were included in the study.
56 % female participants
44 % male participants
Age Distribution: 18-64 years
Spatial Coverage (Country/Region/City): Germany/-/Greifswald
- Proactive recruitment of all clients in the waiting area of the residents’ registration office in Greifswald by study assistants
- A total of 3 x 5,- EUR voucher as incentive (baseline, T3 and T4)
- Advance announcement of follow-up surveys by letter or e-mail in advance
- At least 10 telephone contact attempts per measurement point
- Mailing of questionnaires (paper-pencil or online) with a reminder letter if participants could not be reached by telephone
- Address research via population registers
Sample Size: 1646 participants
- T1: 85% (n = 1407)
- T2: 81% (n = 1335)
- T3: 80% (n = 1314)
- T4: 65% (n = 1074)