I am carrying out research into "some" and "any" using the OEC, which I access via Sketch Engine. Many of my searches into specific patterns with "some" and "any" produce far too many results for one researcher working alone. To overcome this, I have been using the Sample Size Calculator from Survey System (www.surveysystem.com/sscalc.htm) : I set the Confidence Interval (CI) at 4 and the Confidence Level (CL) at 95% and write the total number of searches for the pattern in the Population Size box. So, for example 5804 total examples at CI 4 and CL 95% gives a random size of 544 examples, while 15,000 examples gives a random size of 577. Does this seem a sensible way of calculating random sample size? Does anyone have any better ideas?
I find that frequency searching on the whole sample is helpful, as the random sample will obviously miss something from a larger corpus. http://www.researchgate.net/publication/261946830
Article Is Bigger Better? Corpus and Dictionary Use in the Search fo...
Thank you Phaedra. The frequency search for Sketch Engine, which I had forgotten about, is indeed really useful. However, it will not help me with the type of patterns that I am at present trying to count, as these are functional patterns which are based on my subjective interpretation of the data-e.g. different speech functions such as "expressing desirability" with conditional clauses containing "some". The collocations function from Sketch Engine is a great help in seeing which words typically combine with some and any per se, and with specific searches for some and any in specific patterns -e.g. some and any inside if-clauses, or some and any in wh-questions. However, it does not in itself reveal the most frequent uses of some and any. Other Sketch Engine functions like Word Sketch are also inadequate for my purposes as they only allow searches for nouns (and proper nouns), verbs, adjectives and adverbs. Do you know of any external software that I could attach to the OEC in Sketch Engine in order to search for frequent patterns?
In the absence of such a tool and given the impossibility of physically counting every instance of a particular pattern in a corpus of over 2 billion words (billion being used in the British sense of a million million), it seems to me that the only option available to me at the moment is to use random samples.
Assuming that I will continue using randomized samples, what I would like to know is a reliable way of calculating the sample size needed to ensure that the sample is representative of the corpus as a whole..(Sketch Engine draws its random samples from the whole corpus to avoid, at least partially, the danger of skewed samples. )
I will ask around with people who use search programs. A lot of them are adapted for the Web however. Good luck.
Perhaps you may use AntConc to search for and to identify the pattern in your randomized samples and whole corpus, and compare the results to see whether the randomized samples are representative or not~
- Badges
- Science topic