Which will be a more meaningful way to communicate your results to your audience? This will depend on your audience (some audiences won't know what an ln is) and whether the logged-version is appropriate in your models. You haven't said why you are using logs, but I assume this is because your theoretical model makes more sense with them.

Thank you for the response Daniel Wright

I am using the data for my thesis. A friend after going through my descriptive statistics recommended taking a log of the data. One of the reasons given is that the data contains huge numbers and has skewed distribution features. So the data is log-transformed to make the data look more 'normal' and symmetric.

Attached is the file of my descriptive statistics.

Abdul-Naafik Mohammed , any reason why you want the data to look more "normal"?

If it is not to make the data more meaningful, I'd backtransform (exponentiating in your case) estimates (including the upper and lower bounds of intervals) for reporting your data, but as said above it depends also on your audience. wrt the mean, you might look up the geometric mean.

Alright then.

Thank you once again, Daniel Wright

I think the table of descriptive statistcs should describe (the main features of the distribution of) your data.

The data should be described in its most commonly used form. For most variables this is the sclae it is usually measured in. For some variables, the common scale may already be "logarithmic". For instance, on would report the proton concentration in aqueous solutions usually as pH (what is -log10([H+])) and not in moles per volume.

I usually find the StdDev of little value to describe data. It's often more instructive and less prone to misinterpretation to give some quantiles. The 5-number summary is a good option (https://en.wikipedia.org/wiki/Five-number_summary).

For skewed distributions I would also give the mode. I would give the mean only when this is of particular relevance (it often is not!).

Instead of givin a table you could possibly give (a panel of small) histograms showing the distributions.

The actual data can easily (and should!) be provided as a supplementary table, or as electronic supplement, or made available through a public data repository (https://www.nature.com/sdata/policies/repositories#general). Then there is no need to extensively describe the data. It's available and accessible to anyone who wants to know anything about that data.

Another point:

Don't give too many significant digits. For instance, the row gives a mean = 2.134, sd = 0.258 and N = 379. From this one can approximate the standard error of the mean as 0.258/sqrt(370) = 0.013. So the second decimal digit of the mean is already "in the noise". You should thus only give 2.13. I think even one digit (2.1) would be enough: the difference between 2.1 and 2.2 makes less than one percent or approximately 10% relative error, what is usually quite okay.

I like what Jochen Wilhelm has suggested, although I might quibble a bit over his suggestion of plotting histograms. As the discussion linked below shows, some folks (e.g., Ronán Michael Conroy) are not keen on histograms, because what you see can depend a lot on the width of the bars.

• https://www.researchgate.net/post/According_to_histograms_my_data_is_not_normally_distributed_what_i_do_to_make_it_normal

You could use boxplots or quantile plots instead. HTH.

Fair point, Bruce Weaver , I agree!

This is tangential to the original question, but since that's where this has gone .... A good source on deciding bin width is https://www.tandfonline.com/doi/abs/10.1080/00031305.1997.10473591. Also, important to stress that usually the histogram from a sample of data is meant to be informative about the population distribution. This can affect decisions about bin width. And wrt the original question, the widths of the bins way vary and also the scale for the x-axis could be log, raw, or whatever. But, as I suggested above it seems the original scale will be most useful for your audience.

I'm grateful for the overwhelming responses. It's really educative.

Thank you all

You have to display the original for descriptive But on inferential if the assumption of the model is not satisfied you can transform the data and write as the mean, min, and max value of the transformed data.