Hi
I am working on wind turbine SCADA data sampled at 10-minute interval. I have read it online that we have to perform normality test and if data is not normally distributed then we should use spearman correlation as in Article On the Selection of Input Variables for a Wind Turbine Load ...
But, the recent articles that I came across is applying Pearson correlation without even checking the normality of data. Article Improved power curve monitoring of wind turbines
And it has given good references as well for using Pearson correlation. Even if they have normalised data before performing correlation analysis, isn't that wrong to use Pearson for not normally distributed data?
TIA
It is important to perform Spearman with data not normally distributed.
Ghulam Bilal but the data does not have ordinal variables instead it is continuous.
If the data is not normalised, after normalising it one can go for pearson correlation (this is not a wrong method)
Nadia Masood Khan I think, first check outlier test for data, scatter plot, normality check, levene's test prior to decide to perform Pearson correlation. because Pearson correlation works well when assumptions met. if not you can do it but misleading. Some people use non-parametric alternatives (Spearman rho or Kendall tau) for data has skewed. Some data behave different or irregular distributions and may having non-linear assumptions. it is difficult to plainly say which method suits for your data set. Different data analysis methods and strategies used by different researchers. but my opinion is to check assumptions first, then perform analysis.
Best regards
another strategy is transform data prior to do correlation analysis
Dineth Samarawickrama
Thank you so much for replying to my post. Yes, I have checked the distribution and it is not normal for my data. I did try to perform box-cox transformation but some values are zero or in minus so that's creating a problem as well. For now, I am using Spearman correlation analysis.
Regards
- Badges
- Science topic