(The title is a little misleading, this actually expanded beyond just looking at newbies, to looking at ultras, star sign usage, and class usage over time as well. Also, as Al-Khwarizmi points out, anyone who has finished the game is not truly a newbie -- I just use the term "newbie" in a relative sense.)
There is a lot of opinion on these forums, in the Guidebook, and even in the game itself, about what races and classes are easiest for new players. GE Wizard, for example, is considered very easy, although my personal experiences have differed.
So, I decided to look at the data. I wrote a program to scrape all the YAVP posts in the YAVP forum, and pull out data from those posts which posted FLGs, and also classify them, using the text by the author, as to whether they were first wins or ultras. A lot of this can be improved on technically, but I just wanted to get a rough answer. I got a total of 434 wins, including 59 first wins, and 220 ultras, and 155 non-first, non-ultra wins.
The code used to generate the dataset and figures, along with the dataset itself, can be found at:
https://bitbucket.org/gilesc/adom-analysis
Now, the data!
Race usage for all players
I was surprised Gnome wasn't higher, and to a lesser extent Dwarf.
race.png
Class usage for all players
This one is extremely unsurprising:
class.png
Now, for the more interesting info. These charts show the frequency of a race or class being chosen by first-win players, divided by the frequency by non-first-win players. A higher number means it is more favored by (successfully winning) newbies.
Races preferred by new players more than other players
It seems to me that newbies aren't exactly following the advice of commonly accepted newbie races, except perhaps gnomes. Apparently they like elves, also this figure makes it clear that while seasoned players are using Dark Elves heavily, new players do not like them as much.
first-race.png
(ignore the attached thumbnails below, these were generated before I realized that first ultras should not be counted as first wins)
There is a lot more to be done with this data: R/C combos, ultra win analyses, star signs, and perhaps even looking at what artifacts are good predictors of victories. Also, this code could easily be adapted to look at YASDs, which would also be interesting. I will continue posting as I get more.