We were pleased to learn that our paper is to be accepted for publication, provided we attend to the suggested changes. We have itemized the changes below for your convenience.
Response to review from Donald Reid
This review made no substantive criticisms of the paper, but suggested we consider citing several older papers in British journals that are not indexed on Medline. We have contacted Anne Charlton (Dr Reid’s review contained her email address) in an attempt to gain a copy of the papers, as they are not accessible in the USA, but have not heard back from her.
Response to statistical review
1. MIXOR is an analytic technique that hasn’t been used terribly much in the literature to date, but is potentially very informative for ordinal data. As you will appreciate, the usual run-of-the-mill ordinal logistic regression tests for a constant (ie a dose-response or proportional) relationship between one variable (ban or no ban) and all levels of another variable (five stages of uptake). However, it is entirely conceivable that bans produce a more protective effect early in the uptake process (for example, in the transition from nonsmoker to experimenter) than in later stages (in the transition from advanced experimenter to established smoker) when other factors (like addiction for example) might exert more influence. MIXOR can help to determine if smoking bans vary in their relationship to uptake, depending upon the stage one is looking at.
In the printout of the cumulative logit analysis, there is a test which determines whether the proportional odds assumption for the whole model is met. If the p value is significant, it means that at least one of the variables in the model does not have a dose-response relationship (or proportional relationship) with the outcome variable. More information about this can be obtained from Paul Alison’s Logistic Regression for the SAS System (SAS Institute). MIXOR enables one to see which variables have a non-proportional relationship with the outcome variable, and to enter some as non-proportional and others as proportional, for the purpose of achieving the best fit for an overall model.
A proportional relationship means that there is a constant or fixed relationship between one variable and all levels of the other variable. Thus in Table 2, an "enforced school ban" has a constant or proportional relationship with stage of uptake and the odds ratio is 0.89 across all thresholds. In other words, an enforced school ban decreases the odds of being a susceptible nonsmoker compared with a nonsusceptible nonsmoker by 11%, and it decreases the odds of being an early experimenter compared with a susceptible nonsmoker by 11 %, and so on. For other variables that have a non-proportional relationship with stage of uptake, the odds ratios vary between one threshold and the next. Thus, in the presence of a "total home ban", one can see that the odds of being a susceptible nonsmoker compared with a nonsusceptible nonsmoker are 0.64, whereas the odds of being an early experimenter compared with a susceptible nonsmoker are 0.69. The odds ratio of 0.69 is significantly different from 1.0 and also significantly different from the odds ratio of 0.64 at the first threshold.
We hope this technical explanation has helped. We have simplified our explanation in the text for the audience. We have removed the word ‘proportional’ from the text in the results section since that seems to be the source of confusion. We have explained that the odds ratio refers to the likelihood of transition from one stage to the next, since this appears to be more user-friendly to readers with non-advanced statistical backgrounds.
2. There were gaps in two of the rows because the same odds ratios as appear in column 1 for the susceptible nonsmoker threshold, apply for all other thresholds (ie the odds are proportional or constant). To make it simpler for the reader, we have repeated the odds ratios in applicable rows of the table.
3. We have inserted a Figure 1, which gives an overview of the sampling strategy and response rates. Unfortunately, detailed records were not kept about the frequency with which first, second and third etc reserve schools were drawn; only the overall number of reserve schools approached. The figure shows that 78.5% of original schools selected for the sample agreed to participate, and that 45/77 matched reserve schools (58%) approached also agreed, giving the overall school response rate of 73% (202/277). (As a minor point, the final sample ended up with 202 schools due to late agreement from 2 schools for which substitutes had already been recruited.)
4. We acknowledge that teenage smoking consumption tends to be variable and that the daily consumption measure we have operationalized does not adequately reflect that. We were in the process of constructing a monthly consumption measure to repeat this analysis, since we do have the number of days in the past month students reported smoking, and how many they usually smoked on the days they did so. Unfortunately my research analyst has been away due to unforeseen circumstances and I have been unable to get this analysis run in the short time I have been given to revise the paper. Therefore, to my great disappointment, I have had to cut this section out of the paper.
5/6. We have deleted Table 4.
7. This is a good point - thank you for suggesting it. Concordance between students in schools was high, with over 50% of the schools having 95% or higher concordance (and 80% of schools having 85% concordance or higher) among students in their description of the status of the policy at their school. We have inserted a sentence to this effect in the first paragraph of the results section.
Response to editorial meeting comments
1. We have revised the paper and shortened it to 2,027 words. In order to achieve this, we replaced detail about the sampling of schools with a figure, cut out our preliminary analyses of missing cases, and omitted the section on cigarette consumption, as well as generally made the language more concise.
2. We have revised the results section to assist the audience in understanding the uptake analysis. We have added a footnote to Table 2 to further assist with understanding of the odds ratios.
3. The sampling strategy over-sampled African Americans, Hispanics and students from low-income communities, so that there would be sufficient sample size in these subgroups for inter-group comparisons. We do not believe this is problematic, since our study does not attempt to generate prevalence estimates of restrictions on smoking in different environments and generalize them to the entire US population of school students. Rather, the study focuses on relationships between variables, and we used race/ethnicity as a covariate in each of our analyses.
4. The title now includes reference to the fact that the study is a cross-sectional survey design.
5. We have inserted a sentence in the methods to the effect that students were informed in writing that the survey was voluntary and that responses would remain confidential. The study design and questionnaire were approved by a Robert Wood Johnson Foundation advisory panel with experience in conducting youth surveys on smoking.
6. We have inserted a Figure 1, which gives an overview of the sampling strategy and response rates.
7. Signatures of all authors and a statement of contribution to the study is being sent by post.
Finally, we have formatted the paper to your specifications.
We hope that you will find the revised paper and these responses adequate, so that the paper could be accepted in its revised form. Of course, please do not hesitate to contact me, should further clarification be required.
Please note that I will be out of contact from 11-17 June and 4-19 July 2000. Should communication during these periods be essential, my co-author Frank Chaloupka will be available (fjc@uic.edu or fax 1-630-801-8870) to deal with inquiries.
Yours sincerely,
Melanie Wakefield PhD