Theoretical Dissertation Direction/Resources (need structure)


For many stats based people a theoretical dissertation (e.g., proposing a new methodology) is common. In my institution's department (and field for that matter) theoretical dissertations are not known. I am aiming to undertake this sort of dissertation (specifically a piece outlining and providing examples of graphical display of discourse as a research method) however there is no set procedure in my department for such an undertaking (I may be the first here to do such a thing in the Dept. of learning and Instruction).

I have Googled "theoretical dissertation" and found some loose guidelines.

I figured I ask here as many of you will be familiar with this type of dissertation and may have resources to help guide me and committee. I'm looking for:

  1. theoretical dissertations you may consider exemplary
  2. guides (including institution guides) on writing a theoretical dissertation
  3. general advice

Please use links, citations or general comments to share this information. I'll do the leg work and am interested in what ever people are willing to give/share.

Thanks in advance.


Less is more. Stay pure. Stay poor.

I would also look to ensure that a member on your dissertation committee is familar with these types of approaches. They may or may not be your chair. You may have to get them from outside your department. I can't remember the breakdown of committee members on my committee, but I was definitely required to have a person from outside my department.


Can't make spagetti
most of the dissertations in my field are sort of like that. i don't think i've ever found a set of guidelines as far as how one of these dissertations should look like, but the overall pattern among them is pretty similar.

i will use mine as a motivating example of how i proceed. the main idea was to propose what i considered to be a new (and superior) way of estimating the polychoric correlation coefficient through the use of Markov Chain Monte Carlo (MCMC) instead of the traditional maximum likelihood approach.

the general structure was more or less as follows:


(1) overview of what is currently being done in the literature. so, in my case, some of the history behind the polychoric correlation coefficient. how Pearson came up with it and used series to estimate it. the problems that arose with that and a general description of how things are done now by using maximum likelihood. there's always a criticism at the end of why there is a need to improve on the current state of affairs (in my case, the usual stuff: ML needs large sample sizes, the possibility of non-positive definite polychoric correlation matrices, ML is bad at handling unbalanced frequency tables, etc.)

(2) overview of the statistical theory i would rely on to improve this in general (so quick intro to Bayesian statistics) and in particular (the Albert 1992 paper that describes the Gibbs sampler that i was going to use). also, my own predictions of why i think the method i'm proposing will improve upon the standard method (so maximum likelihood)


description of the simulation studies to test whether my claims about the new method being better holds or not. lit review your stuff to make sure you're tapping on the usual simulation conditions from previously-published research: sample size, correlation size, discretization points, skewness/kurtosis of the latent variable, etc.


what it sounds like (the results of my simulation)


so it's basically just saying "here's how people do things. here's who *I* think things should be done and here's the evidence to show it.

to be honest with you, the only thing i'm proud of in terms of my thesis is the Appendix. it was initially going to the be crux of the whole thesis but my advisor deemed it to be too technical for the other reviewers in my committee so it got relegated all the way to the back :(