I am not statistician, however sometimes I have to use different programs to build graphs. Does it make significant difference using modern programs in stats such as R, and others, or Excel makes almost similar operations? Some of my Mathematician friends explain that the basis, for all statistical programs is the same, therefore there is no significant difference which program do you use. The patterns and figure will be almost similar. Can anyone answer little bit in detail?
As a general answer, simple statistical calculations should be the same in any program, whereas the outcome for more advanced models may be different among programs.
However, even for common statistics or tests, there may be differences in the methods used or default behavior of the function. For example, there are different ways to calculate the 95% confidence interval of the mean. It is important to know which method is used, and importantly if the method is appropriate for the data. As another example, for a chi-square test of association, one software package might apply a continuity correction by default, and another might not.
Often there is not a obviously correct way to do a certain test. For example, R will use Welch's t-test by default for a two-sample t-test. Some argue that the best approach is to use Welch's t-test routinely. I believe SAS reports the results of the [non-Welch] t-test and the the Welch's t-test and then leaves it up to the user to decide.
My Dear George
The answer of Salvoters is agood answer about your question,all methods or software give the same results in statistical analysis only different in options and the output screen but the basices are the same
with best wishes to you and Salvoters
Dear Salvatore
Thank you very much for your answer. I'll write you soon about the matter and wish to discuss some results with you.
Cheers
George
Dear George
The different Softwares give you the same result for the same data set, but the different statistical procedures may be given different results according to their sensitive and accuracy.
Good Luck
George,
You can, sometimes with a lot of work, get to essentially the same result with almost any software. In principle a Turing machine can do the job. However, some software are more helpful than others. For example, W Cleveland in his book, 'The Elements of Graphing Data', sets out principles of effective graphing, based on research on visual perception for the purposes of clear and precise communication of information. These are built into base R and several R packages, as the defaults. The defaults in Excel by contrast are hardly ever in conformity with these principles. One has to wrestle Excel into submission.
But people rarely do, and Excel graphs have exerted an unhelpful influence on the quality of graphs (inefficient use of space, needlessly low resolution, low information to ink ratio, graph type inappropriate to the data type,...) that we find in applied science journals.
R can/does take some work too. Having a good foundation for a graph, more often than not you have to work on font sizes and labeling of axes etc. for publication, and one needs to learn R code, which may not be realistic for old-bugger scientists.
So use Excel for graphs if you like, but first learn what constitutes a good graph, and don't accept what Excel dishes up.
A similar story with stats, but here I am much more wary of Excel, its limited breadth of options, its lack of transparency, its ability to scramble data. Use a stats package for stats. If you are doing agriculture, GenStat would be a good choice.
Wishing you well
Murray Hannah
Using Excel for scientific work requires either a serious disdain for precision requirements for publication (e. g. what does Excel computes exactly ? Is that the "right" solution to your problem ?) or a fair amount of masochism (i. e. "programming" Excel to give you exactly what you need, i. e. in most cases reinventing a (difficult to build) wheel that's already available in R or Stata). And don't get me started on graphics (Salvatore and Murray make good points, but there's much more to it : see Cleveland's book recommended by Murray).
Yes, one can get publishable results from Excel, but that entails much more work than learning how to use R (or, probably, stata) ; furthermore, this work has to be reproduced on the next paper, etc...
And I didn't even allude to the possibilities offered by "reproducible research" tools such as Sweave or knitr : IMNSHO, those tools are *indispensable* for creationg a reliable paper. If \LaTeX is too hard for you, at least try Markdown (which can be used by knitr) ; including your analysis/graphics code in the text of your report/paper fees you of any concerns about publication consistency. As far as I know, the (primitive) abilities offered by DDE/OLE/COM won't allow that.
You should be aware that, historically, Excel has not had a good reputation as statistical software: see for example http://www.pages.drexel.edu/~bdm25/excel2000.pdf . This could lead to different answers from Excel and other packages for exactly the same problem/data. While some improvements have been made to Excel statistics-stuff as new versions arrived, these have not always seemed to clear all the problems: see http://www.forecastingprinciples.com/files/McCullough.pdf .Thus you should at least ensure that you are not using an extremely old version of Excel, and possibly search for later versions of these assessments of Excel. See also https://www.researchgate.net/publication/266025083_On_the_Accuracy_of_Statistical_Procedures_in_Microsoft_Excel_2010
Article On the Accuracy of Statistical Procedures in Microsoft Excel 2010
Dear Colleagues
Thank you very much for your comments, it helps a lot to make right choice.
I look forward to hear more comments.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS