The key to developing an excellent, university-level (or rather university-acceptable) research paper, I have found, is to begin with sources. My own personal method for developing a clean paper on the quick is to:



1) Get at least 1/4 of the paper's length in direct quotes or paraphrases into a text editor, with citation data.

2) Write in why each study is important.

3) Write in how all of the studies combine to for one generalized outcome.

4) Sort the citations by similarity of why statements and then generate topics, to organize and outline the paper.

5) Write, read, and edit out errors.



This simple pattern is the A pattern, I have found, but it has still never been enough. It is difficult to take confidence in my reflections on data, however convincing it may be to a professor or thoughtful it may seem to a laymen against these methods....
I think that the real problem is that I randomly commit type I and type II errors without even knowing due to an inability to clearly calculate and summarize datum weight. As a result of these negative assumptions, I often over-write and add a few extra pages to my research or spin expensively on examples at the end of each study explanation....



Graphing data first not only creates a clear and concise summary of the data for each individual study, but it also allows me to weigh the data against each other and come to a more clear, more accurate interpretation of data sets, no matter how small or large, the first time.



Prior to today, this paper may have been my favorite kind for understanding somethng new: http://www.ori.org/~keiths/bibliography/statistics-meta.html --the old method, a horribly one-sided review that corners the reader into a maximum of three axes and stone-walls them into a perhaps erronious assumption.



Sensitivity graphs, like forest plots and regressions to adjust for multiple outcomes, time-points, comparisons, and subgroups; to adjust for continuous moderators; or to find breaking points for better research proposals. I think that the key is, being able to plot a x-bar/y-bar range for as many similar outcomes as one can find, associated with as many hypernyms as possible, and then find the easiest and most difficult to answer questions. Also, being able to review old research and see how accurate it was is nice. For those of us who cannot help but to ask our readers questions.... Why not try one more method by which to answers these questions, ourselves? Especially questions like: "I can't be right, can I?" To which the answer is almost certainly yes and no.



To learn more about meta-analysis, go here: http://www.pitt.edu/~super1/lecture/lec1171/index.htm.