I'm trying a cancer survival analysis, and as part of my analysis I'm looking at the demographics of my population and whether they differ statistically.
While I'm looking into statistical difference between age at diagnosis and race of the patients for ex., I noticed that using a chi-square test (with bonferroni correction) to compare Age at diagnosis (in categorical increments of 15) there was NO statistical difference.
However, when I use ANOVA test with Age at diagnosis as a nominal variable (i.e. the exact age in number) there was a statistical difference.
I'm no expert in statistics, I believe this data manipulation decreases the validity of results. I'm not looking for a significant p-value but I would like to know why did this happen, and which value to report since both variables looked valid to me.
The simple answer: is you change your data, the results will change.
As you say "I'm no expert in statistics" - why do you torture your data and misinterpret results of statistical analyses? It is absolutely fine to not being an expert in statistics. But in that case, please cooperate with a statistician. There is much more to consider than can be explained in RG. Eventually, you will benefit, your research will be of better quality.
I am not an expert in cancer medicine (although I know some background and facts, and I think I understand some principles - just like you may know some things about statistics) - it would never come into my mind that I would go and give diagnoses to patients! I would surely involve an expert in cancer medicine / diagnosis (if my work would require to come up with such diagnoses).
Dear Jochen Wilhelm,
Thank you, but I would value providing some references for this particular issue much more.
Of course when it comes to data that's being published a statistician is always involved. I'm using this data set to practice and learn survival statistics and data analysis on my own, Thus I'm having doubts that I'm sharing here.
Regards.
Some good recommendations are given in the attached two papers. I overall agree with Jochen however. Best, D. Booth
Interesting question Mohammad, in fact there is a huge debate about the validity of many articles due to data manipulation and alteration from researchers hunting for a significant p-value. Hence journals started asking for data sets and recruiting statistician to review the analysis carried out by researchers.
As you pointed out, this is data manipulation and misinterpretation that definitely decreases the validity of results. The main question you need to ask is what is the purpose of your research? are you looking for the change of risk with every year of life? or are you comparing multiple categories of age? and if it's the latter, what did you base your categories upon? why did you come up with increments of 15s instead of -for example- 10s? you should not come up with questions based on the results, it's the other way around.
To publish research you certainly have to seek the help of an experienced statistician, but for you, I would recommend the attached paper.
All the best.
Thank you very much Majed.
Indeed you're correct. I'm very new to Survival statistics (still learning the basics) and the example in my question is just a general reference to multiple errors I noticed in my analysis due to poor variable definition and not defining the research question properly.
Appreciated. Regards.
I would also caution the interpretation of what the P value actually means. Attached is an article describing some practical applications and interpretations of statistics FYI (though does not go into detail about the survival methods you are exploring).
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA