Soc 388, notes on the final paper assignments.


For 2007




Pct of final grade

Paper abstract

Tuesday, November 6


Paper Draft

Thursday, November 15


Final Paper

In class Thursday, December 6





1) Abstract:  The abstract should be 1-2 pages, and should present the data and a couple of key questions you hope to answer.  The abstract should include a description of where the data is from, and the full dataset itself, in the form of one or multiple tables. 


2) Paper Draft.  The draft paper should be 10-12 pages. Paper drafts (along with cross tab datasets and stata logs) must be submitted electronically, preferably in MS Word format so that the TA and I can comment on them and return comments to you, and also keep a copy for ourselves. This paper draft should focus almost exclusively on your research questions and your data analysis which attempts to answer those questions.  The first step is, before you get in to loglinear models, show some tables of simple statistics.  Describe your research question in simple terms of counts, percentages, change over time, or whatever simple measure is appropriate. Describe the number of cells in your dataset, and the total unweighted count of observations. Then generate a series of models to test your hypotheses.  Use tables (as in homework 2 and homework 3) to summarize your models.  Be clear in the text about how you are defining your dummy variables, and be clear about which statistical tests or variables help you answer your key questions, and why.  Tables need to be MS word tables, in real English.  You may need a brief literature review to frame your research questions, but this should be no longer than 1-2 pages.  You will be asked to submit an electronic version of your dataset via email.


3) Final Paper.  The final paper should be 12-20 pages, and should reflect your responses to Professor Rosenfeld's and TA's comments from the draft.  Literature review should be no more than 5 pages. The intended style is journal article style, so if you are having doubts about how to frame your argument or how to present the data or your results, check the syllabus and read some of the recommended readings.

            New notes on the final paper:

            * With your final paper, you must also turn in your draft, with Professor's and/or TA comments on it, so that we can judge your progress.

            * Data sections must clearly describe the full dataset (including number of cells and total unweighted count), source of the data, the exact definition of your variables (including how many levels each variable has), and what kinds of cases are excluded (if any).

            * As in the draft, the first part of your analyses must present simple tables with simple statistics. Don't get into the loglinear models until you have shown some simple tabular analyses.

            * Final papers must have all the usual parts such as an abstract, and a full bibliography

            * Every paper must have a table that compares various models and their goodness of fit (like the homework).

            * In the table listing the models, pay attention to goodness of fit by LRT. In other words, you need to push your data to find at least one model that fits reasonably well (P>0.0001 is one arbitrary line- 'reasonable' P might be smaller if the dataset and number of cells was very large) by the LRT compared to the saturated model. This model may be too complicated or may not converge well, so you may choose to focus your discussion on simpler models, but you do need to show that you can go through the rigorous work of fitting models to the data.

            * Think about creative ways to present (in tables or figures or 3D figures) either coefficients from your models or coefficients from a saturated model.

            * All tables must be Microsoft word tables that you make, with headings that use real English words, and so on (in other words, you can't just copy from the STATA log).

            * The key to a good paper is your ability to frame a question, generate hypotheses related to that question, and test those hypotheses in sensible ways with the models.