spunky

well, we're being led by a professor and his students who are interested in seeing how the narrative around educational policy has been changing since around the 70s until now. something about the laws changing to favour the rich over the poor (no surprises there, lol) under Bush I and Bush II. what i found more appealing (because i honestly find politics a tad bit boring) is the approach. the two quants in our group who do text analysis are coming up with the matrices of measures so that the other SEM modeler out there, with me, can fit a latent growth model. my guess is that once all the text analytics are gathered (mostly classical latent semantic analysis) they'll pass their resultsto the other quants (us) who will fit the latent trajectories to see how these concepts (like "fairness" or "equality" in education) change over time. i was drawn to it because i found the appeal unique. i have never seen people try to do SEM on text analyses but, once you start thinking about it, it makes perfect sense. just as people who can do meta-analysis also using SEM or multilevel models. what are you doing IDA for?

well, in my opinion, if you're already doing IDA chances are you're already pretty advanced in your stats. besides, with the advent of this overload of info, people who are proficient in meta-analysis, IDA, big data analytics, etc. are a hot, hot asset. i'm currently part of a research team that does text analysis in educational policy (so tracking and measuring trends among thousands and thousands of pages from published articles related to it) and we have the equivalent of 11 King James Bibles worth of content. who's ever gonna read through that!? QUANTS TO THE RESCUE!!!!

well, your first article (De la Torre) deals with Item Response Theory (a *must* in Psychometrics for Education but not necessarily as popular among Quant Psychs), dwelling particularly on Bayesian estimation. the Bayesian approach to data analysis is super hot-hot-HOT now among social sciences but it can be opposed by hard-line frequentists, so it really depends on whether the person you'd like to wrok with is a frequentist or Bayesian. a good friend of mine who works with Herb Marsh (one of the best in our field) once said to me something along the lines that if you submit a manuscript for publication with the words "Bayesian" or "Markov Chain Monte Carlo" on it, your chances of publishing go up significantly among quant journals. it turned out to be true for me, so i jumped in that bandwagon. it may not be yours, but you can always say you have an "interest" in Bayesian modelling. you may want to say something about having an interest in Simulations (as in Monte Carlo simulations) and you're eager to learn new software that enabales this (people like Mplus or R which is rapidly becoming, as the New York Times phrases it the "lingua franca" of data scientists). you could say that maybe you have some project that uses longitudinal data and you're interested in either using multi-level models or structural equation modelling (SEM) to tackle it becuase you've heard it's the correct way to go about these things. i like the Kim article you mentioned where they use Multi Dimensional Scaling because i don't see that used very often for longitudinal data analysis. i'm more familiar with it being approached from a latent growth curve perspective. i can't really help you with the other two articles because it seems like their more assessment-oriented rather than stats-oriented, and that's not my area of expertise. just keep in mind that you are not expected to know ANY of this stuff. you're going to gradschool to learn it, plus it depends on the prof you may want to work with. even though my uni has a small Quant Psych program, for instance, the profs in it are diametrically different in the type of grad student they want. Prof A is happy to get people who are enthusiastic (and obviously numbers-savvy) because he'll teach them the rest. Prof B doesn't like to take in anyone who doesn't have any previous programming experience and a BSc, which guarantees said candidate took at least some math beyond Grade 12. in general, you just have to show that you feel comfortable-enough around Statistics to learn more about them. overall i'd maybe just touch on three main points: ( a ) you're interested in latent variable models (structural equation modelling, item response theory, etc. all fall under this umbrella) ( b ) you're OK with expanding your knowledge as far as software goes... maybe in going beyond GUI-heavy (Graphic User Interface) programs like SPSS, Minitab or JMP and working more in a syntax/programming-only environment ( c ) you're open to learn about statistics. everything else will fall in its place.

Lisa44201 is right. why don't you maybe give us a reference of some article of these professors to get an idea of what kind of models you're interested in? when i wrote my statement i just lumped everything together in having a "keen interest in latent variable modelling" whatever that was supposed to mean. it got me in so i guess it can't be that bad? lol

if these are your research interests then ditch Quant Psych and go for Ed Psych all the way. I've worked in the quantitatively-oriented educational measurement program from my University and i'd say your research interests are just right in line with the assessment people rather than the quant analysis people. and for programs as such, you don't need to have a heavy background in Statistics. i think with your usual upper-level methods stats course should be more than enough. but if you've audited a doctoral course AND TA'ed an undergrad stats course i'd say you've demonstrated more than enough proficiency in the area to be a qualified candidate. all the math/stats/programming experience you need you'll learn along the way.

i thought i was gonna take a year off before starting gradschool... ... truth is i ended up enrolling on a 2nd bachelors degree while i waited for my applications to be processed. couldn't live without academia for more than 4 months

would switching to a more qualitative-methods-based program work, maybe? or not switching programs but just switching your approach to research? i can see that both political science and psychology (my field) have very quickly become heavily quantitative programs and tend to look down upon theory that is not backed by solid empirical research. being an absolute pragmatist, i'd advise you to, honestly, stick with it. if you resist the "question-hypothesis-method-data gathering-analysis" process, you're also closing the doors to publishing in many reputable journals. which also closes the doors to advancing your career and diminishes the chances that you'll do any "rewarding teaching" once you're done.with your PhD. remember the motto: "publish or die" (where "publish" implies question-hypothesis-method-data gathering-analysis). and if psychology works anything like political science (and i have a feeling it does, in this case), people tend to resist publishing theoretical articles by newly-minted scholars. not sure why they do that, but i'm sure they do.

yes to both.... they even cover my husband. but then again this is Canada

Structural Equation Modelling :-)

just as a minor updated, i did end up getting a TA position! thanks for everyone who helped give me the courage to apply!

you know, you actually place a very intresting question here and i think you hit the nail right in the head with it. when i started programming back in high school i wasn't very good at it (and, heck, even after these many years i know i have a long way to go to get better) but i think you just need to develop two sets of skills: how to think sequentially and how to think modularly. force yourself to structure your thinking. what do i need to get from A to B? what are the series of actions? and thinking modularly in terms of being able to organise series of lower-level commands into big bundles of computer behaviour. how to do that? well practice.practice.practice and patience.patience.patience. learning how to code is no different from learning another language. there is a syntax and there is a logic to it. but we don't get it unless we use it and use it and use it over and over again. it's interesting to see how as society becomes more dependent on technology, more and more peolpe are actually less and less inclined (and less interested) in how this technology works and how to interact with it. i praise initiatives like in code.org where big names in the computer world (or even in the entertainment one like Will I Am from the Black Eyed Peas) try to promote this idea of getting middle schoolers and high schoolers coding as soon as possible. worlds of oportunity open up for you. heck, even my colleagues who do QUALITATIVE (yes i said it! such a dirty word ) research marvel at how fast you can go through coding texts for content anylsis with just a a few computer programming tricks. i have to say it pains me a little bit to see brilliant PhD students and professors who should be using their knowledge and talent to interpret data spend endless hours of boredom doing unnecessary, repetitive tasks such as capturing data on a spreadsheet. a good friend of mine to whom i've taught R told me that the best way for her to learn it was to do all her homework assignments and class projects in SPSS and R. yes, she mentioned it took her twice the time as everyone else and it was frustrating as hell... but with every new assignment she bacame faster and savvyer until the point now that she has jumped in the R bandwagon and ditched SPSS altogether. i think sometimes people miss the point that ever since SPSS (and SAS along with it) went licence-exclusive, being familiar with only those software packages hinder their ability to market themselves outside big, well-established academic institutions. what if you start working in a small community college with not enough funding to pay for your SPSS licence? or if maybe you want to help some NGO who's already under some severe budget constraints? or even if some journal editor asks you for some alternate analysis that is not available in the drop-down menus you're used to? no one is born known how to code or how to do statistics... but we can do something about it if we try :-) oh! and UCLA's website for everything-statistics is a life-saver: http://www.ats.ucla.edu/stat/

for a nice, concise introduction to GAMs, you can always read Chpt 12 of Extending the Linear Model with R by Julian Faraway. i've found Faraway's book to be a pretty useful resource to become familiar with more advanced regression-like, linear modelling techniques. it gives you a little bit of theory, then some toy dataset and explains each bit of the theory through examples in R code and how to interpret the output. if this is the first time you've dealt with GAMs, it's probably a good place to start. in general, the "little red books" (everything from the Texts in Statistical Science collection) are pretty nice refernces to have around. the ones i've worked with use R code but i'm sure there's plenty of ones with SAS code out there.

well, i guess the first couple of things that come to mind is (a) why do you need for it to be non-parametric? (don't get me wrong. non-parametrics are cool but i've sometimes found there's usually an easier, more robust parametric alternative out there) and ( when you said 'non-parametric' GLM... which one of the non-parametric family of methods are we talking about? local smoothing? generalized additive models (GAMs)? lasso regression?, etc...

well there you go. VBA (and apparently SPSS syntax by extension) are procedural programming environments. R behaves more as an object-oriented programming environment (although it isn't strictly speaking object-oriented). people here in the Sciences prefer it more because you can jump from it to Python, Java or C++ much more easily because they are all object-oriented programs as well. what about working in SAS instead of R? SAS is procedural as well. it might be more in line with how you're used to thinking about programming. i mean, there's nothing wrong to try and learn both approaches but i guess it could save you some endless hours of googling, lol.

make that very hefty more of a **VERY VERY VERY HEFTY** tuition. i'm talking from experience here. and there are programs that won't even let you enroll unless you're guaranteed some sort of funding... which can be tricky for international students since funding is usually given first to domestic students. yes, this is also me talking from experience.

tinn-R is definitely something good to have around. and i think it's very commendable that you're attempting to dwell more in R to make sure you can get a better hang of it. which programming environments are you used to? C++ maybe? Java? just wondering here what your programming experience has been like to see if you're more familiar with procedural programming or object-oriented programming in terms of paradigms. usually, the paradigm in which you were trained in (or trained yourself) sort of defines the kind of programming "grammar" (like you said) you're used to. and the weird thing about paradigms is that they don't necessarily overlap so if you're thinking along the lines of the "grammar" of a certain paradigm, you may be completely off when trying to work on a different one. and i totally agree with you. programming in SPSS is a mystery to me. i know SPSS syntax is a spin-off from like this, really, REALLY old version of lingo but, it doesn't behave like lingo at all so i have no clue what it's doing. that's kind of why i ditched it altogether, heh.

well... it was just deduction (oh i'm such a Sherlock Holmes ). i've been studying Bayesian Statistics since i was doing my BSc (i did a double major in Statistics and Psychology) so i'm familiar with most of the textbooks on Bayesian Statistics out there (there aren't many). Kruschke's book is probably the *only* one that is geared towards people who do not have a mathematics background, and it is specifically aimed at social scientists. i know your technical training did not come from a Statistics Dept because, if it had, you would be so intrinsically familiar with R that you could even code it in your sleep (as i think i mentioned previously, R's the default in Statisics, Comp Sci, etc.). and when you mentioned "The book we were using had code to go with the examples, but the code didn't actually work in the program" my mind went like this: "which textbook on Bayesian Statistics is geared towards non-Statisticians/non-Mathematicians who are not familiar with R and caters to social scientists?" the answers where either Kruschke's book or Lynch's "Introduction to Applied Bayesian Statistics and Estimation for Social Scientists". but Lynch's book is too technical and has very few well-developed examples in R code, so the only logical conclusion left was Kruschke's now that we have estabilished you used Kruschke's book, i can say a few things. Kruschke makes a very decent effort in trying to make an overly complicated subject approachable to people outside from Stats/Math. the problem is that he tries to pack too much in his book. he barely devotes... what? only the first 3 chapters to something that even remotely resembles the theory of Bayesian statistics? and then he throws the reader into example-code-example-code mode until the end! honestly, i agree with you. the code in the book is buggy but that's mostly Kruschke's fault (not R's). he creates this expectation that if you were to just copy-paste his code into your R console and run it, it would run. but it doesn't! packages get updated all the time. and... well... his code *IS* buggy! now, if you're already good with R that isn't too much of a problem because you can debug what Kruschke wrote pretty easily. but if you're not already familiar with R, i can only imagine, as you mentioned, the endless of hours annoying boredom googling the cryptic error messages R spits back at you. my main beef with Kruschke's book is that instead of trying to teach you the logic behind statistical computing for Bayesian statistics and how to build your own MCMCs, he just goes and says "here's the Bayesian alternative for the t-test. here's how you do it in R. copy-paste this code, change the variables and you're ready to go", which is an approach which simply doesn't work for Bayesian analysis. the problem is, of course, that trying to teach said logic would imply the technical/mathematical details would have to increase and that would probably turn off a lot of people not only form his book but from Bayesian Statistics all together. i guess i would just say maybe you'd be willing to give R another chance, coming from a different starting point? really, once you've mastered it, it'll happen just as with Mplus: you'll never look back. heck, you can even ditch Mplus altogether!

i'm assuming this was Kruschke's book?

noooo!!!!!! don't miss out on R!!! you're gonna get left out from what the NYTimes is calling the "lingua franca" among statisticians and data analysts: http://www.nytimes.com/2009/01/07/technology/business-computing/07program.html?pagewanted=all&_r=0 what was your issue with R and Bayesian statistics??... because my master's thesis was... sorta both of them. a Gibbs sampler for the polychoric correlation coefficient..... done in R. from scratch, lol. MATLAB i like but it's expensive :-(

i gotta say i'm a R fan so i stick to the lavaan package and if i wanna get fancy i use OpenMX. Mplus is pretty awesome but since it's becoming the go-to software for people doing SEM it's starting to undergo it's own little "SPSS-ification" process and that sucks big time :-/ plus its graphing capabilities are... well... let's just call them "mediocre" at best :-)

i think it does depend on the types of analyses you'll be doing. AMOS can be very (and i mean **VERY**) limited in the types of models you can fit.... but then again this is coming from a quant person and we're known for working around with complicated stuff... sometimes a little too complicated lol

i kinda miss it to be very honest with you....

you do have to admit, thought, that this one summer has been particularly warm and relatively good for beach-related activities!!!

thanks you both!

thank you everyone for your kind words. i think you're all right. i want the teaching experience and i think i deserve it, particularly after 2 yrs of letting everyone take up all the opportunities. i guess after being constantly bombarded by news about student debt this and student debt that and how hard it is for young people to get jobs and this and that i thought i was becoming part of the problem if i applied for positions that others might need more than i do ... however ... heck, i need the teaching experience! i'm gonna be filling out and sending my application tomorrow. if someone ends up without a TAship well too bad then. it's my turn to play (lol)