In publishing researches it is important to report main assumptions of statistical method employed -including tests and concrete data worked-. Method is the way you handle the elements used to produce and interpret your information and arrive to your conclusions. At the end you obtain some model that must fulfill not only your dataset, but also those main assumptions of method. Some language with its inner rulls must be used to express and publish them.
It is important because it is related to logical coherence of any research. If the fundamental assumptions are questioned, perhaps it means that the research may be valid for the set of assumptions used, but not for other possible sets of assumptions. Methodological premises and procedures should be key elements to be observed by science partners when replicating experiments, and by science educational authorities. I think it contribute to creative progress in research.
I think it's important to report statistical models they have used as a help for other researchers, and I believe that a good journal must be careful about that.
Yes, it is good practice to always check that the assumptions of parametric tests are met before running the test and this should form part of your report. There is also the robust use of parametric tests: some parametric tests can work reliably even when their assumptions are not satisfied. This leads to the need to satisfy weaker assumptions.
When I do a stats test (for example the t-test), I am testing if the assumptions are valid. That's why I report the assumptions. So I report if my data is normally distributed, although "NON-NORMALITY has almost NO EFFECT ON P-VALUES when we compare means, especially when the sample sizes are moderate" by the central limit theorem, which says that as the sample size increases the distribution of means approaches the normal distribution.
I also report if the variances for the two treatments are equal, but if I have equal N for both groups, the alpha level is not affected. And I report the sample size of each group. So it looks like I am justifying that it's ok to use the t-test, since the assumptions are valid. (I learned quite a lot from my 2 very HUMOROUS stats profs in varsity, while I was doing Masters and PhD...)
In publishing researches it is important to report main assumptions of statistical method employed -including tests and concrete data worked-. Method is the way you handle the elements used to produce and interpret your information and arrive to your conclusions. At the end you obtain some model that must fulfill not only your dataset, but also those main assumptions of method. Some language with its inner rulls must be used to express and publish them.
It is important because it is related to logical coherence of any research. If the fundamental assumptions are questioned, perhaps it means that the research may be valid for the set of assumptions used, but not for other possible sets of assumptions. Methodological premises and procedures should be key elements to be observed by science partners when replicating experiments, and by science educational authorities. I think it contribute to creative progress in research.
You see all statistical analysis has its own requirement regarding type of data, data scale, data distribution etc. You need to know this because failing to meet the assumptions would caused your analysis to be invalid. Thus, you need to report that each and every requirement (assumption) is met.
I agree with all the comments above and I think I can add a few words. Statistical tools create their tools starting from mild assumptions: for example, there is a sample of independent variables with common normal distribution. From this, point estimators, intervals and hypothesis testing are built. Therefore, these tools work well if the initial assumptions are true and that's why it's important to check them and report them. If the initial assumptions are not met, inferences can still be made, but its validity may be doubtful.
Hundreds of papers published during the last fifty years make it clear that violating standard assumptions is a serious concern. Wilcox (2012, Introduction to Robust Estimation and Hypothesis Testing, Elsevier) summarizes why and describes methods for dealing with known problems. Several other books deal with this issue. Testing assumptions turns out to be ineffective. These methods generally don't have enough power to detect situations where there is a problem. The only known method that is reasonably effective is to try a modern robust method and see what happens. There is now a vast array of improved techniques.
The Importance of Testing Assumptions Before Running Statistical Analyses
Thursday June 14, 2012
Elite Research explores the importance of testing statistical assumptions. Elite Research is a global provider of research design and statistical consulting. They support academic, corporate, medical/health, and non-profit researchers in designing, collecting, analysing, and reporting efficient and accurate results.
Many statistical tests have assumptions that must be met in order to insure that the data collected is appropriate for the types of analyses you want to conduct. Common assumptions that must be met for parametric statistics include normality, independence, linearity, and homoscedasticity. Failure to meet these assumptions, among others, can result in inaccurate results, which is problematic for many reasons. When testing hypotheses, running analyses on data that has violated the assumptions of the statistical test can result in both false negatives and false positives, depending on the particular assumption violated.
Most, if not all, statistical software packages, such as SPSS and SAS, do not automatically check these assumptions; rather, they assume that these have been met as they are conditions in which the logarithmic functioning of the program underlie. Therefore, researchers must thoroughly explore the data and run appropriate preliminary analysis to ensure that the data does not violate the assumptions of the statistics planned to be used.