The choice is highly dependent on the data. Let representing the data in the most illustrative, and most informative way be the objective. Consequently, a comparison between alternative graph types will help you decide the suitable one for your data. The log graphs are suitable for representing data of significant differences in the values of the points.
If you mean distribution and parameters of distribution, the argument you presented is true. However, if question is related to use of data, I agree with Mohamed EL-Shimy.
Both are important and each is very useful. The raw data is the source of the information, but raw data can be bulky depending on nature and population handled. Transformation comes in during the analysis. Some raw data may hide information that is revealed in the transformation. Transformation or analysis of the data brings out the juice which leads to conclusions.Each has a part to play and should not be ignored.
The transformed data emanate from the the raw data. It is often a product of adjustments simply to bring out the flavour needed to get the job done. Both of them are very important and should be used in data description.
One can ask, '' Parent and child who is more important?'' There will be no transformation without raw data. In geology, it is like from Field work to Laboratory analysis. One starts the race and the other gives the finish.
[The promised comment to my previous comment] A publication is (at least in principle...) written for readers, namely for competent readers potentially willing and able to evaluate the published results and eventually use in their own work. This is usually possible only if the data are presented in generally interpretable - what in practice almost always means the least transformed - form. So, presentation of e.g. direct measurements are in this respect by far the best, if this is impossible then simplest statistics (mean, SD) are preferable, while e.g. coefficients of correlation, to say nothing of principal components or the like, are practically worthless for researchers not studying the same questions, applying the same methods, using the same transformations). This was e.g. my problem with potentially very interesting publication whose conclusions were hardly believable and I strongly suppose they resulted rather from inappropriately transformed data than from real anomaly - but without raw data I am unable to either prove or disprove this (those eventually interested may see the attached publication).
Raw data provides the most accurate representation of observations and offers flexibility and transparency. Transformed data simplifies complex patterns, normalizes data for analysis, and reduces noise. The choice depends on analysis goals and data characteristics; raw data is often used for exploration, while transformed data may be necessary for specific tasks like predictive modeling. Both have their place in analysis, guided by the requirements and desired outcomes.