Alpha diversity correlation: To rarefy before analysis?

I am running an alpha diversity correlation (spearman for normal variables otherwise pearson) on soil nutrients. Should I rarefy the table before this correlational analysis as you would for beta diversity? I haven’t found any reference to this in the forum or in my analysis of ecological communities book.

Thanks any advise.



Hi @insectnate!

Generally yes, if you are using simple measures of alpha diversity like Observed OTUs, Faith’s PD, Shannon, etc. you’ll need to rarefy as those simple calculations are assuming that an even sampling has been done of the data (which isn’t true for amplicon sequence data). Another option would be to model the sequencing depth as part of your correlation, but QIIME 2 doesn’t have an option for that and you should probably have a statistician on hand to help model that correctly.

If you used something like q2-breakaway you could avoid rarefying entirely as it can account for the uneven sampling effort in your data. However it does need singletons to make this work, so you may need to adjust your upstream processing a little bit for that. (cc @Amy_Willis, @Pauline_Trinh)


Hi Evan,
Thanks so much for the help! I’m using exact sequence variants from the dada2 pipeline so singletons aren’t a problem.
I was looking at this part of the qiime alpha diversity tutorial and I noticed this entry
Could I used this plugin function to accomplish the model estimate of missing taxa due to undersampling?
Chao1 confidence interval: Calculates chao1 confidence interval
Confidence interval for richness estimator, Chao1
–p-metric: chao1_ci

Chao1 index: Calculates Chao1 index
Estimates diversity from abundant data
Estimates number of rare taxa missed from undersampling
–p-metric: chao1

Hi Nate! Chao1 is a valid estimate of total species richness only when all species are equally abundant – unlikely to be the case! breakaway is a good tool.

@Pauline_Trinh wrote an awesome tutorial so do reach out if you have questions or feedback. breakaway’s models are pretty flexible and statistically rigorous so gives error bars, too, of course! In the R version of breakaway, we also have a method for hypothesis testing that uses those standard errors – I encourage you to check it out.



Hi Amy,
Thanks so much for the help and advice. I installed it without error and am attempting to run breakaway as a qiime2 plugin but I am getting a cryptic error message…
I opened an issue on github and posted the log file that was generated.


This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.