# How to figure out the behind concept of statistical-look problems?

The following is my experience doing some researches linked to statistical concepts. In the most situations I had no idea on how to organize the questions, how to ask for answers, and how to provide feedback showing my acceptance or rejection. I'm certainly not talking about the language issue but pointing on that it was not clear to me what the concept behind my question is. Sometime it comes out that there is a misunderstanding regarding the question. Some people do not tolerate the lack of formality (~standards) in the question and so their replies are not useful but discouraging. I believe that there is some sort of kindness and helpfulness trying to answer questioner with the hope being satisfied.
Provided this background, I would appreciate it if you would provide me a list of: tools / tips-and-tricks / methods / skills / techniques / strategies / ... to find out how to seek for solution / help / ... in facing with statistical problems.
For example this site (Stats.StackExchange.com) is a tool providing a forum based technology to ask statistical questions. Some people probably recommend reading books etc which I consider them not a shortcut solution although for young scholars it could be a good idea. The reason is the questioner in my question has lots of thing to do and also has other profession.
My experiences showed that asking a clear/complicated question from an academic (unknown/famous) who does not know you has no way to be succeed. For a keen researcher it is however disappointing.
For this question I tried to be clear as I could if you see it not clear enough, you got the problem. Any ideas?

I accept that the main question (the idea behind this post) might be a bit hard to be caught up by immediate-commentators. Briefly, in the first time facing with a problem we would have some guesses about the related topics/subject/methods in statistics that could help us to understand the problem and to propose solutions. This post discusses the problem solving strategies to connect a proper statistical analysis to the problem being investigated even roughly.

Although productivity could be developed by the help of professional statistician however this is not the case to be addressed in this post. The main purpose of this post is to help self-learning practitioners who have enough background about the main concepts being investigated however do not have a clear idea if there is any modern/classic developed concept regarding the problem or not. For statistics as for other parts of sciences there are always tips-and-tricks that could help the researcher to speed up the progress of productivity, at least I have a bunch of them.

I would recommend commentators to give a fair opportunity for discussions even if it is not their favorite. Please think positive and deep you'll see good points in this post. Nevertheless, to me what you comment or mark or vote will be useful even if those are not directly addressing the question and regardless the tone/language all are appreciated.
@cardinal: Thanks. the provided link is an interesting post.

• Fastest and easiest way would be to hire a professional consultant. – Roman Luštrik Oct 25 '11 at 10:27
• @Roman Personally, I cannot see any benefit in this comment. For some reasons people are very active to disturb instead of providing positive and productive thinking. This comment itself can be considered for the purpose of smile (the original idea of the commentator, I guess) however it opens a door for non-professional acts such as down-voting etc. In addition, commentators are advised to know what the purpose of the comment is. Destroying a huge building with a small dynamite is not big deal. Try to build a small house then you'll enjoy so much. – Developer Oct 25 '11 at 12:03
• It's actually just incredibly hard to understand what you are on about. Your previous question seems to be pretty much the same, but it's easier to understand: stats.stackexchange.com/questions/17407/…. Maybe you could have another crack and try to get to the point. Maybe this belongs on a more culture/language-based stackexchange site? – mdsumner Oct 25 '11 at 12:12
• I just added a comment in the question. Hope it could add a clarification to the original point of the post. I'll learn even if the comments and answers go in a way that looks unsatisfactory to me. My idea is to solve a problem that would be of interest of many non-professional statistics consultants. – Developer Oct 25 '11 at 12:39

I can only talk from my experience and I totally agree with Joris. The key is "what do you want to find out from your data set?"

I am not statistically trained and I have been asking all sorts of questions on correlation analysis because I am very clear in my mind that I am testing bivariate relationship i.e. I am interested in finding the nature and extent of the relationship between two variables. Nothing more (to my peril, perhaps!).

Some of the responses to my questions have been very confusing but many have been excellent. If anything, the responses have prompted me to think more clearly about my method of choice which is Pearson's r. (Sometimes the best method is non-statistical i.e. commonsense approach.)

I am still struggling to understand why some contributors to my questions think that Pearson r is very basic (see here) but at least I have been alerted to the notion that there are better or perhaps more robust ways of answering my questions.

My advice is to use this site to your advantage. Ask clear questions and put in as much details as possible. And hope a professional statistican notices your question.

I did not learn much about statistics in high school/university, partly because my teachers had little idea what they were talking about. I hold a masters degree (by research!) but I have gone back and educated myself on basic concepts such as standard deviation. The point here is that learning is life-long. (if all fails, use YouTube!!)

No one can give you a complete answer. Only you can put the pieces together to address your questions to your satisfaction. Thankfully, you can find some of the missing pieces here on this site from people who are so generously sharing their knowledge.

I cannot thank my contributors enough!

Keep using this site!

• (+1) Thanks for a good advice. Your experiences are valuable to me. For some of my questions I never got a satisfactory answer (well, so far). But for some others I got amazing complete answers. This means that we can contribute to the community by questioning but finally the community decides how to answer to our questions. Therefore grateful to your nice advice the best is to get benefit as much as we can from the given answers and kindly efforts. Behind anything my appreciation goes to the time being spent viewing, answering and commenting my posts. – Developer Oct 25 '11 at 14:49

Roman gave you the best advice possible. In addition, I'll take a shot at it. But remember :

Learning statistics is like learning any other thing : There. Are. No. Shortcuts.

1) know what you are talking about. It's no use to ask questions about concepts you don't know. So indeed, reading a book and -even better- following some courses to get a thorough understanding of the basics is a necessity.

2) Once you have a problem, check that it fulfills following requirements :

• it is defined (about a particular case)
• it can be described in at most two short paragraphs
• it is isolated (so no "I have this dataset whadoIdonow?")

3) If all the above is done, take it to somebody at your departement/faculty/whatever with a thorough background. Discussing problems face to face is always better, as it is more interactive.

4) If the problem persists, you can ask a question here following the guidelines in the FAQ. They will tell you all of the above as well by the way.

In any case, if you fail to understand this, you will be reminded about

fortune('brain surgery')


I wish to perform brain surgery this afternoon at 4pm and don't know where to start. My background is the history of great statistician sports legends but I am willing to learn. I know there are courses and numerous books on brain surgery but I don't have the time for those. Please direct me to the appropriate HowTos, and be on standby for solving any problem I may encounter while in the operating room. Some of you might ask for specifics of the case, but that would require my following the posting guide and spending even more time than I am already taking to write this note.

-- I. Ben Fooled (aka Frank Harrell) R-help (April 1, 2005)

EDIT : some further explanation of my point

You're a researcher and you have a dataset and want to find out which techniques there are to tackle this problem. That's what your question is about. But :

1) before you formulated an hypothesis and can formulate what you want to know, forget searching. I get people at my desk daily asking me "what would you do with this dataset?" and my first question is always "What do you want to find out?". In more than 90% of the cases, the asker fails to answer this question. Stop looking and start thinking.

2) If they answer the question, I often get an answer like

• "find a relationship.".
• "Between...?".
• "I don't know, what can you get out of it?"

Wrong answer again. How many people ask a question like "I have this huge dataset with this and that counts and 400 other variables and I want to summarize this data." That's not a question, that's a lack of understanding the concepts of data mining. If you can specify that you want a score indicating the amount of ... and have a scientific reason to do so, then we're talking.

3) Check the amount of papers in statistical journals. You need somebody who's following this in order to know which methods are available to date. And I kid you not, I've been studying Generalized Additive Models (one small method) for two years before I understood all aspects of it. So you need experts. In one of my consultancies, the firm asked to deliver a 2-day course for applying my models to the new data. After I explained them about model building, model checking, spline bases, circular splines, penalty versus no penalty, confounding and a bit more, they understood that even a 2-day course by a trained statistician is not enough to do this correctly.

4) SE has some professionals around. They are willing to help you out with troubles when applying a method. Reason why I, as a professional, answer only few questions here, is because :

• I don't have time to waste on questions where it's obvious the OP has absolutely no idea of what he's doing.
• I don't have expertise in all available methods, and I tend to shut up about things I don't know enough.

I hope this clarifies my answer, and also clarifies why the approach you're asking for will be impossible to find. There are no shortcuts.

• Some parts of your answer is good but some is irrelevant. Not sure what you understood from the question! – Developer Oct 25 '11 at 12:06
• I read again and again your answer and finally I realized that your answer has nothing to relate to the concept/question which I put on discussion. I'm afraid that you completely missed the points in the question. A proper language is good for answering but a proper purpose is much more needed, my friend. Think positive and you'll find lots of good ideas in my question, hopefuly. – Developer Oct 25 '11 at 12:10
• @Developer : I'm a full-time statistical consultant, and I know very well what your question is about. I just fear you don't understand my answer. Without understanding of statistical concepts you shouldn't even start asking questions. If you understand the necessary statistical concepts, Google, SE, wikipedia, textbooks, ... all will do. – Joris Meys Oct 25 '11 at 12:21
• @Developer Although it is true that your question is not directly answered in this thread, the community is sending a strong, clear message in its upvotes to this reply and its voting on several comments. – whuber Oct 25 '11 at 14:06
• @Developer: The first link in this answer may provide some helpful information and perspective related to the "There. Are. No. Shortcuts." remark. – cardinal Oct 25 '11 at 16:51