I don't want to talk too much, but I have read that there are some goals/tests for which a which a normal data distribution is not important, and others for which it very definitely is. And my suggestion would be, if a book did exist: put that and similar information on page 1 of every chapter, in a small heads-up box. TMN On Fri, May 2, 2014 at 11:58 AM, Narandra Dashora <narandrad(a)gmail.com&gt; wrote: -- Best regards, Timothy M. Nall Assistant Professor National Quemoy University Kinmen, Taiwan

Discussion of assumptions is a question of econometrics and is not sofware specific. But, may be, some of me experience would be interesting. I modeled  well behaved (in the other senses) cross-sectional data for regression with power-law disturbances. (I admire Taleb, but don't agree with him on some of his thoughts on normal distribution) At the standard levels of 0.05 and 0.01 t-stats behaved well even for sample size of 20. The distortions are in the tail. Say, a program (gretl, for example) gives p-value of 10^-30, as true p-value of power-law distributed t-statistics is 10^-5. Such distortions are critical for value at risk (it deals exactly with the tails!), but have no consequences on decisions on significance. From my point of view, the most serious problem is that many economic time series are in fact time non-homogeneous. --- Оригінальне повідомленн я --- Від кого: "Narandra Dashora" < narandrad(a)gmail.com > Дата: 2 травня 2014, 07:00:34 If it is not undue intervention as an humble student of Econometrics ( I am 70 years old ) I feel The various assumptions about  Normal Distribution are hardly met in real  life data. If one has sound theory yest there is no co integration of data or unit root is present then there are two golden ways. A. Increase the sample size B. Reexamine the theory Incidentally I may   comment that the concept of spurious regression has been oversold On Fri, May 2, 2014 at 8:54 AM, Tim Nall < tnall.ling(a)gmail.com > wrote: All, Again please refer to previous disclaimers about the fact that I certainly and truly have no idea what I am talking about w. respect to statistics, and perhaps with respect to other issues as well. In an earlier post, someone or other expressed a concern that gretl might be  perceived as a toy program that is purely for educational purposes, and not for serious research. I have no idea whether or not that's the case, but I would suggest that focusing on pushing gretl as a serious tool is not the only path toward your ultimate goal. Me personally, I would recast a perceived shortcoming as a serious strength: push it as an educational tool. The way to do so, as I mentioned earlier, is by writing a book. The book should be data-centric and outcome-centric. That is, the traditional teaching approach would be to think, "I have x number of statistical topics to cover, and they can be ranked in terms of difficulty and inheritance of concepts from one another, so I will present them according to that rank. I will discuss the math and theory first, and perhaps (or perhaps not) tack on a skimpy example in the end." This is a forest-for-the-trees approach IMHO. Me personally, i would approach the book as "You have (this) type of data, and you want (this) type of outcome, so you should use (this) type of model, unless your data has (these) conditions, and the way to deal with (these) conditions is (chapter)." I can truly embarrass myself here by offering my own recent exp. as an example: in my very modest research, a set of 10 OLS regressions (on time series data) that were truly beautiful and perfect in every way (they conformed very, very precisely to my initial set of hypotheses) sat on the pages of my document for months before I realized that the results were spurious due to non-stationary data. [Econometricians and statisticians can politely refrain from giggling.] So ch. 1 of your book, rather than immediately presenting the nuts and bolts of OLS, could first present the same sort of case (very quickly). And so on. Make top-level ideas (such as which problems to check for before considering any given approach) *very* easy to find at a glance. Got math? It goes in appendices. And here's the point: you would want your book to help gretl catch on as an educational tool, but sprinkled throughout the book you could also mention (w. brief details) its ability to do serious research. No harm done. Then if it catches on in the former context, eventually (lag a couple years) people will pick up on the latter idea -- also realizing, hey, you know, it's free. So that's all I have to say. Thank you for your patience. TMN _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users

On Fri, 2 May 2014, Tim Nall wrote: Heh. I'm not quite sure myself of what my ultimate goals are ;) But allow me to explain (briefly) why I don't quite agree. First: there already several books out there that use gretl as a vehicle for teaching basic econometrics. One is the OUP book by C. Dougherty, but the most popular resource on the web is arguably Lee Adkins' excellent companion to "Principles of Econometrics" (which I guess has been downloaded a gazillion times by a gazillion undergraduate students). More in general: I'm not particularly interested in world domination. My goal, as a member of the gretl community, is primarily in providing the economic scientific community with a nice free (as in speech) tool for doing econometrics in the best way possible. Of course, "doing" econometrics means teaching it, but also using econometrics as an intermediate tool for applied economics (maybe for research, maybe for other purposes) and even developing new statistical techniques. So, what I'd like gretl to do is to fulfill the needs of the widest possible spectrum of users, from the undergraduate student who struggles with OLS to the theory geek who needs to run a comprehensive Monte Carlo experiment to assess the finite-sample properties of tests for nonlinear cointegration in VARFIMA system with alpha-stable, infinite variance disturbances (the example is totally made up, but doesn't it sound cool?). You'll agree with me that it's not humanly possible for a small team of volunteers to achieve all this, for a multitude of reasons. So, we need to expand the pool of contributors to the project as much as possible. But knowledge of C is, understandably, rather rare in the profession; and besides, even if you can code very well in C, finding you way through the source code of gretl is no mean feat, given the dimension the program has taken over the years. So, if we want the project to have a future at all, we need to push hansl as much as possible to attract new people into the development and to make it possible to them to transfer their econometric skills into code easily and efficiently. This is why we are, nowadays, working primarily on hansl: to ensure gretl has a future. And besides, because it's fun! ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti -------------------------------------------------------