46

I was wondering if there is a statistical model "cheat sheet(s)" that lists any or more information:

  • when to use the model
  • when not to use the model
  • required and optional inputs
  • expected outputs
  • has the model been tested in different fields (policy, bio, engineering, manufacturing, etc)?
  • is it accepted in practice or research?
  • expected variation / accuracy / precision
  • caveats
  • scalability
  • deprecated model, avoid or don't use
  • etc ..

I've seen hierarchies before on various websites, and some simplistic model cheat sheets in various textbooks; however, it'll be nice if there is a larger one that encompasses various types of models based on different types of analysis and theories.

Andre Silva
  • 3,070
  • 5
  • 28
  • 55
dassouki
  • 1,219
  • 1
  • 17
  • 25
  • 4
    +1, however, I'm mixed about this sort of thing. Often they seem to exist so that someone can not know much about the analyses in question, but still scroll through the list, find a name that meets their conditions and then run through the procedures. In short, I fear they lead to 'cookbooking' data. In addition, I suspect they reinforce the idea that these are distinct tests w/o an underlying continuity, & that the test (p-value) is all that's important. IE, they help to solidify misconceptions & conceptual biases about statistics. Nonetheless, they do have some value... – gung - Reinstate Monica Oct 28 '12 at 21:22
  • 2
    I'm not downvoting this, but I want to reinforce @gung's warning. Any such list will be totally misleading, unless several of the sections are multiple pages long per technique (Caveats, When to use, When to not use, etc), and I can predict that several suggested sections will inevitably be misleading (Expected variation/accuracy/precision, Has it been "tested" in different fields, etc). This overall list will be a step backwards for science. IT COULD be useful to have a list of deprecated techniques (with replacements listed), but... – Wayne Aug 04 '16 at 14:30

5 Answers5

22

I have previously found UCLA's "Choosing the Correct Statistical Test" to be helpful: https://stats.idre.ucla.edu/other/mult-pkg/whatstat/

It also gives examples of how to do the analysis in SAS, Stata, SPSS and R.

lucidyan
  • 113
  • 5
Thylacoleo
  • 4,829
  • 5
  • 24
  • 32
  • 1
    A simple help through the rseek.org or r-help mailing list will get you on your way with dealing with most (all?) of the methods in R (which is the program package I would suggest to anyone). Good link. – Roman Luštrik Aug 06 '10 at 07:47
  • 1
    R examples were added to the website! – Drew75 Mar 22 '14 at 07:00
  • 2
    this page seems to have moved to https://stats.idre.ucla.edu/other/mult-pkg/whatstat/ (as of 2017-09-20) – Andre Holzner Sep 20 '17 at 07:12
21

Do you mean a statistical analysis decision tree? (google search), like this (only with extensions): alt text
(source: processma.com)

?

BTW, notice that the chart in wrong in that the tests it offers for median are not for median but for rank... (it would be for median if the distribution is symmetrical)

Tal Galili
  • 19,935
  • 32
  • 133
  • 195
  • 2
    Something similar to that but includes more information rather than just the name of the test. We have some of these charts in urban and transportation modelling. They'll show a large table where they specify tests per type of problem. They also list caveats, expected time/duration, input and outputs, etc – dassouki Aug 04 '10 at 17:35
9

Reading "Using Multivariate Statistics (4th Edition) Barbara G. Tabachnick" I found these decision trees based on major research question. I think they are quite useful. Following this link you'll find an extract of the book http://www.psychwiki.com/images/d/d8/TF2.pdf see pages 29 to 31

tosonb1
  • 321
  • 1
  • 2
  • 6
  • I assume you have the book, what's in chapter 17 (it's referenced in the document you provided) – dassouki Aug 05 '10 at 11:41
  • Chapter 17 covers: 17 An overview of the GLM 17.1 linearity and the GLM 17.2 bivariate to multivariate statistics and overview 17.2.1 bivariate form 17.2.2 simple multivariate form 17.2.3 full multivariate form 17.3 Alternative research strategies I hope this can help Regards – tosonb1 Sep 03 '10 at 02:01
  • the link is dead (404 error) – Mehrad Mahmoudian May 30 '18 at 10:03
8

Here is a collection page: http://sasdataguru.blogspot.com/2011/05/online-statistics-cheat-sheet.html

Jason
  • 51
  • 1
  • 1
1

Since when is regression an hypothesis test of anything? If by"regression"why is meant is curve fitting or correlations (pair-wise or multiple) the only "test" is between some relation vs. no relation. Figures like this own their origin to Siege's l956 book.

user321272
  • 1
  • 1