WB parametric or non-parametric - (Jan/09/2012 )
I've been using ANOVA with Bonferroni for statistics on my western blots (in published papers), but I've got an evaluation before defendig my thesis questioning that you can use that instead of a non-parametric test with western blots, especially since n=3 and you can't tell if the data has a normal variation or not. Any suggestions to what I can answer if the question comes up?
After reading up on statistics (why did I do this first now...), I definately see her point, but at the same time I see plenty of papers using ANOVA for WB, so I can't be completely wrong? or are researchers just clueless when it comes to statistics
In general people are clueless about statistics; I frequently see incorrect use of statistics in published papers and in other media such as in news reports. It is a sad fact that there is very little knowledge of the correct usage of stats, even amongst senior scientists. I wish there was more teaching of the correct usage of stats for non-statisticians (i.e. us poor biologists and other scientists).
With an n=3 you should have used a non-parametric test, and/or looked at your data to see if you can assume a normal distribution of the results. ANOVA is not really useful for samples below about n=30, and the power diminishes rapidly below that. Unfortunately there is not much that you can say about incorrect usage, other than to follow up and make corrections where needed!
bob1 on Tue Jan 10 00:35:22 2012 said:
In general people are clueless about statistics; I frequently see incorrect use of statistics in published papers and in other media such as in news reports. It is a sad fact that there is very little knowledge of the correct usage of stats, even amongst senior scientists. I wish there was more teaching of the correct usage of stats for non-statisticians (i.e. us poor biologists and other scientists).
With an n=3 you should have used a non-parametric test, and/or looked at your data to see if you can assume a normal distribution of the results. ANOVA is not really useful for samples below about n=30, and the power diminishes rapidly below that. Unfortunately there is not much that you can say about incorrect usage, other than to follow up and make corrections where needed!
I certainly agree that statistics is a subject many researchers in the life sciences put into the voodoo category and just do what other people do before them, even if the conclusions are doubtful to say the least. Concerning your problem, your evaluator is right in questioning your assumptions that are prerequisites to ANOVA (namely normality), but honestly I would doubt any test (even the non-parametric alternative to ANOVA, i.e. the Kruskal-Walis test) that compares groups of three.
bob1 on Tue Jan 10 00:35:22 2012 said:
In general people are clueless about statistics; I frequently see incorrect use of statistics in published papers and in other media such as in news reports. It is a sad fact that there is very little knowledge of the correct usage of stats, even amongst senior scientists. I wish there was more teaching of the correct usage of stats for non-statisticians (i.e. us poor biologists and other scientists).
With an n=3 you should have used a non-parametric test, and/or looked at your data to see if you can assume a normal distribution of the results. ANOVA is not really useful for samples below about n=30, and the power diminishes rapidly below that. Unfortunately there is not much that you can say about incorrect usage, other than to follow up and make corrections where needed!
according to my stats-book, ANOVA actually can be used for non-parametric samples as well when n is over 30. dosen't say anything about its usefulness in lower numbers. I think I'll just have to agree with him and point out that the variation is so low, that any test would find them significant.
I've found a paper where the main evaluator used ANOVA in his WB-results with n=4, so maybe I'll have to point a finger at at him if I'm being attacked to hard (carreer suicide-option)
I had a quick question for you on this. I've just gotten into doing more western blots etc and was wondering how you go about generating statistics when comparing samples obtained across different blots. I alwasy run a "no treatment control" on all of my blots, so I can express everything as % control. However, if I want to do a statistical comparison to my control group I cannot because it is, be definition, set to 1 so there is no way of calculating any sort of error associated with it.
I've seen one place that suggests having a sort of "standard sample" that is run on all gels that can be used to normalize everything (including the control). Is this standard practice when needing to generate statistics on samples run across blots?
Thanks in advance.
Having standard samples is common practice, as is normalizing the data to a standard control such as beta-actin.
Having said that, in my experience, trends, but not absolute numbers are comparable between blots, making it very very difficult to get any meaningful statistics out of WBs.
Thanks bob1. Well I wouldn't try to get any sort of absolute numbers, but I presume using a standard sample on all blots would allow for the generation of relative quantification compared to a particular control, correct? That way everything can be expressed as % control...