Data analysis methods and applications

Predictors of the Transition From Experimental to Daily Smoking Among Adolescents in the United States

By reviewing the methodology and analysis sections of the study, I discovered that the quantitative method used was a complex sampling design with restricted-use data. The authors of the study used data from the National Longitudinal Study of Adolescent Health as their focus, so they had the opportunity to study a large amount of data without being required to collect it from survey participants or interviews (Park, Weaver, & Romer, 2009). Examined were strata at the individual student level and schools as clusters, as well as the subjects and their unique selection criteria (Park, Weaver, & Romer, 2009). There were 134 schools and 90,000 students participating. Each of those students filled out a 45-minute questionnaire regarding expectations for the future, their friendships, and their health (Park, Weaver, & Romer, 2009).

I found that the covariates, which were demographics, PBT-related, and non-theory related factors, were measured at the baseline (Park, Weaver, & Romer, 2009). These were then used to be predictive of the smoking status that would be seen in these individuals at the end of the first year they were followed (Park, Weaver, & Romer, 2009). The smokers who were selected as the sample were between 12 and 18 years old, and where experimental smokers when they were selected for the baseline (Park, Weaver, & Romer, 2009). Only those who also provided outcome information at the end of the first year were able to be included, as that outcome data was critical to the determination of the study (Park, Weaver, & Romer, 2009).

Otherwise the analysis could have not used the multiple imputation procedure that was expected for this study. The study ended up with 4,903 smokers at baseline (Park, Weaver, & Romer, 2009). Males were 48% of the sample, with more than 52% of them classifying themselves as white or Caucasian (Park, Weaver, & Romer, 2009). For the definitions of smokers -- experimental and daily -- CDC guidelines were used. Those who smoked cigarettes "all 30 days" in the past 30 days in the collection of year one data were considered daily smokers, with others being considered only experimental smokers (Park, Weaver, & Romer, 2009). Many factors were measured, including religion, self-esteem, friendships, connectedness to family, and marijuana use, along with exercise, seatbelt use, and grade point average (Park, Weaver, & Romer, 2009). These factors were theory-related, and there were also non-theory related factors addressed.

Specific Steps Used for the Analysis

I discovered that there were specific steps used for the quantitative data analysis in this study. These began with the design of the complex sampling. SAS Version 9.13 was used to actually analyze the data (Park, Weaver, & Romer, 2009). The smokers at year one were studied to see whether they were experimental or daily, and a multinomial logistical regression analysis was done in order to find influential predictors that move someone from smoking as an experimental behavior to becoming a daily smoker over the course of one year (Park, Weaver, & Romer, 2009). That allowed three categories to be analyzed, but I determined that there were also three other reasons why multinomial logistic regression was used. These other reasons were incorporating such a complex sampling design, examining multicollinearity between all the covariates, and managing any missing data with multiple imputation. By taking all those steps, better conclusions could be drawn by the authors.

Level of Significance

The concept of "level of significance" is important because it provides a likelihood that the sample chosen for the study will not be an accurate representation of the current population (Denzin & Lincoln, 2011). If a level of significance is very low for a study, that would indicate that the authors are highly confident that the results can be replicated (Gorard, 2013). For example, if the level of significance is 0.05, there is a 95% chance that the study can be replicated, and if the level of significance is 0.01, there is a 99% chance that the study can be replicated (Franklin, 2012). Testing this level begins with the null hypothesis, as that is where many quantitative studies begin (Franklin, 2012). The 0.01 and 0.05 significance levels are the two most commonly seen levels in the majority of educational research (Denzin & Lincoln, 2011).

Significance Level for the Current Study