6 Top questions from Editage's recent webinar "Unpacking Data Distributions"
We hope you were part of the webinar Editage conducted on the topic “Describing and understanding data distributions” on February 28. Dr Jacob Wickham, Managing Editor of the journal Integrative Zoology and Assistant Professor at the Institute of Zoology in the Chinese Academy of Sciences, provided expert insights on the importance of data distribution in statistical analysis, various types of data distributions, measures of central tendency and variability, and the best practices for real-world applications. The event was attended by hundreds of researchers around the world.
If you missed the webinar or want a quick recap of the most-asked questions by the participants, keep reading!
We bring to you the top 6 questions asked by researchers like you during the session along with Dr. Wickham’s responses. If you’re interested, you may watch the entire webinar here.
Q. What were the significant differences between parametric and non-parametric statistical analyses?
Dr. Jacob Wickham’s response: What you’re looking at between the parametric and non-parametric statistics is whether the assumptions are being followed. If you’re doing a parametric statistical analysis like a t-test or an ANOVA, you are assuming the data follow a normal distribution and that there is a homoscedasticity and heterogeneity of the variances. So, if those assumptions cannot be followed, then you should switch to non-parametric statistics. So, before you start analyzing your data, you can do the Shapiro Wilke’s test and also the Levine’s test to show if your data follow a normal distribution and variances are equal, because these are some of the assumptions of the parametric statistics. Don’t worry if your data are not normal. You can switch to a non-parametric test and do your statistics using non-parametric tests and you still get your work accepted by scientific journals.
Q. Are there any good websites that serve as a ready reference for understanding fundamental statistical concepts?
Dr. Jacob Wickham’s response: I would look at Dr. Shapunov’s website. It gives a general overview of different statistics and why you should be choosing certain statistical test over another. It also gives an overview of all the main concepts. There’s also a couple of books like “Biometry.” This is a general statistics book in biology. This is a good book for understanding basic statistics. And then, there’s another one titled “Applied Multivariate Statistical Analysis.” These are statistics books that I was using as an undergraduate, master’s student, and a PhD student. One great thing is that statistics doesn’t change. It’s been a field in itself and applies to biology, chemistry, engineering, social sciences. These are great references.
Q. When should we transform data for experiments conducted in laboratory conditions?
Dr. Jacob Wickham’s response: Sometimes, you can do a data transformation if your data follow like a logarithmic distribution or an exponential distribution; there’s data transformations that you can do and you can have your data follow a more normal distribution. In some cases, this is explained in the statistics textbooks where you can transform your data prior to a data analysis.
Q. When should we go for a Bonferroni correction in pre-clinical tests? Please give an example.
Dr. Jacob Wickham’s response: Sometimes when you’re doing statistical tests, let’s say you didn’t do a proper ANOVA and you conducted about 10 t-tests and overused t-tests. In this instance, it would be good to do a Bonferroni correction, where you take the p-value and divide it by 10 because you did 10 t-tests where you should have done an ANOVA and this is a proper way to get around overuse of t-tests. In my doctoral dissertation, I compared 3-4 different groups and the number of combinations of testing two different treatments. So, I calculated the number of combinations of comparisons you can make, say 12. I took 0.05 as the p-value and divided it by 12. That was my new p-value, much lower but I was able to show that the data were significant.
Q. When will you decide to use type 1 or type 2 error?
Dr. Jacob Wickham’s response: Your Alpha at 0.05 takes into account both type 1 and type 2 errors. If you have an alpha value that’s very small, let’s say 0.01, you might actually have a difference in the means between the two populations but you can’t detect the difference because your alpha value is too small. 0.05 is really a compromise between type 1 and type 2 error. I just want you to realize that both of these types of errors exist.
Q. Should we use non-parametric over parametric tests based only on the failed normality or homoscedasticity tests?
Dr. Jacob Wickham’s response: That’s not always true. If you have categorical data like a Wilcoxon rank, it is a non-parametric statistical test in itself. Also, I think, chi-square is non-parametric as well as it is also based on categorical data. So, sometimes beforehand you’ll know that you’re using non-parametric statistics and you don’t always have to start with these tests on normality and homoscedasticity.
Looking for expert advice from a statistician in conducting your next research project? Editage’s Statistical Analysis & Review service can help. Book your consultation today.
Comments
You're looking to give wings to your academic career and publication journey. We like that!
Why don't we give you complete access! Create a free account and get unlimited access to all resources & a vibrant researcher community.
Subscribe to Conducting Research
