Halliday & Matthiessen (2014: 510, 510n):
… all combinations of taxis and mode of projection are not equally probable; while some are very frequent in text, others are relatively rare. As the numbers in Table 7-17 indicate, out of 1,392 instances of projection nexuses in a sample of texts from a range of spoken and written registers, there were only 15 instances of paratactically projected ideas; in other words, 97.5% of all 595 projected ideas in the sample were reported rather than quoted. In contrast, projected locutions were more evenly balanced between quoting and reporting in the total sample – around 46% and 54%, respectively. The sample contains more projection nexuses from spoken texts (845) and from written texts (557); but it is still interesting to note that in spoken discourse paratactic locution (quoting: 53%) is favoured over hypotactic locution (reporting: 47%)*, whereas in written discourse it is the other way around: paratactic locution (quoting: 39%) is significantly less common than hypotactic locution (reporting: 61%).
* In spoken casual conversation, this difference is even more marked: quoting accounts for two thirds (65%) and reporting for one third (35%). In written news reports, the ratio is almost the reverse.