What explains the difference of a de facto larger frequency of vowels of one writer compared to another? In the statistics data I examined, a vowel had higher probability in the text from the female Swedish authoer compared to a Russian male author. The statistics I cite compared the male and female use of consonants and vowels indicated that the probability of next sound being a vowel was much higher for the Swedish female author compared to a Russian male author. The probability of next sound being a vowel and the probability of next sound being a consonant could be explained to vary by style, by book, by author, by language and/or by gender (male/female)

Making statistics on material either women or men wrote, I hypothesize that there are more vowels when the writer is a female and more consonants when the writer is male. Are there any evidence for or against my notion? Did anybody make a study like that? Does it have any purpose besides being a "fact"? A purpose I can think is revealing forgery when a man for instance in a text pretends to be a woman or vice versa, a woman writing to you pretending to be a man then according to patterns you could get an indication.

Edit: I changed it to a real hypothesis about how sounds change since we may wish to compare phoneticallly if doing a real study that could indicate for instance whether the next message is from a man or a woman.

Edit: The statistics say there is a statistical difference between 2 books specified as the markov matrix for if the next sound is a vowel or a consonant given that the current value is a vowel or a consonant.

  • 1
    Making statistics and it seems to me do not go very well together... Did you do a correct statistical analysis or not?
    – nico
    Aug 14, 2011 at 13:55
  • 4
    Maybe you can cite these statistics? Voting to close until you do.
    – Robusto
    Aug 14, 2011 at 14:35
  • 1
    Is this just your experience, or do you have a source that makes this claim?
    – Jay Elston
    Aug 14, 2011 at 16:52
  • 1
    I guess the same study with specific words would give much more interesting results. Aug 14, 2011 at 17:08
  • 1
    @Niklas: So you're saying that a larger sample of authors might not change the result? I beg to differ. A sample size of 1 per group is what's called "anecdotal" evidence.
    – Robusto
    Aug 15, 2011 at 12:33

1 Answer 1


Just out of curiosity I have done some quick statistics.

I downloaded the following books from Project Gutenberg

Men writers

  • Alice's Adventures in Wonderland by Lewis Carroll
  • Adventures of Huckleberry Finn by Mark Twain
  • Moby Dick, or, the whale by Herman Melville
  • The Adventures of Sherlock Holmes by Sir Arthur Conan Doyle
  • The Picture of Dorian Gray by Oscar Wilde
  • Paradise Lost by John Milton
  • The Works of Edgar Allan Poe — Volume 1 by Edgar Allan Poe
  • War and Peace by graf Leo Tolstoy
  • Dracula by Bram Stoker
  • Treasure Island by Robert Louis Stevenson

Women writers

  • Secret Adversary by Agatha Christie
  • Jane Eyre by Charlotte Brontë
  • Frankenstein by Mary Wollstonecraft Shelley
  • Pride and Prejudice by Jane Austen
  • Sarah Orne Jewett
  • Ramona by Helen Hunt Jackson
  • Home Influence by Grace Aguilar
  • Middlemarch by George Eliot
  • A Season at Harrogate by Mrs. Hofland
  • Wuthering Heights by Emily Brontë

After removing the common Project Gutenberg header, I've read the files in R, split them into characters and let it count vowels and consonants.

I had a total of 8725700 characters for men and 11468186 for women

Here's a graph with the ratios consonants/vowels1 calcolated per book (showing mean +/- standard deviation)

Consonants/vowels ratio

There is no statistical significance in the two groups (p=0.89, t-test)


I played some more with the data and I got this bargraph of usage of the single letters.

usage per letter

Again, you can see no major differences between men and women writers

EDIT2: I repeated the analysis with 10 books per group. I would say that there is definitely no difference

1 I considered a, e, i, o and u as vowels, the result does not grossly change including y.

  • The difference in use of q is interesting, but running the same test on 26 samples and finding one of them to have statistical significance (even if q does, which I didn't check) is not unusual, and likely not statistically significant. But IMO it's worth looking into with a larger sample: try a whole bunch of books and look for differences in q use. (Of course, if one of the characters in any of those four novels by women has a q in his name, then that explains everything.)
    – msh210
    Aug 14, 2011 at 16:52
  • I haven't done the stats for the single letters. Anyway running 26 t-test would be wrong. I am not sure which is the best test to run here, as the variables are dependent on each other, probably some sort of linear model would do. I'll try and add some more books later ;)
    – nico
    Aug 14, 2011 at 17:06
  • Well, running tests is probably unnecessary, as you can see the proportions are about equal — except for q. So maybe add more books and just test q. Do any of those four books have a character with a q in his name? Wikipedia doesn't list any in the articles on the novels.
    – msh210
    Aug 14, 2011 at 17:11
  • 1
    @msh: I repeated the analysis with 10 books per group and now you can see no difference at all.
    – nico
    Aug 14, 2011 at 19:45
  • 1
    @Niklas: well, sure, pick any couple of the books I've chosen and you're likely to see some difference. But if you only take 2 books there are too many disturbing factors, such as the nationality, gender and age of the writer, the use of slang, the type of writing (journal article, scientific writing, kid's novel etc) that you would not take into account. A proper analysis has to be done on a larger corpus I would say. But I am not an expert in text analysis...
    – nico
    Aug 14, 2011 at 19:56

Not the answer you're looking for? Browse other questions tagged or ask your own question.