Database comparison - sample length based on word or utterance
Published at: 2016-02-22
Why recommend using total number of words to control for sample length when comparing a client to the database?
Comments
Posted on 17/02/2016 14:40:11
I would like to know why you recommend using total number of words to control for sample length when comparing a client to the database. I have always used number of utterances as the length-control factor (e.g., a 50 utterance sample compared to samples 50 uttererances long in the database). That way, I can use TNW, NDW as points of real comparison between my client and the database. Thx.
SALT Support
Posted on 11/03/2022 20:24:31
If you control by number of utterances and the client's MLU is shorter (or longer) than the database comparison samples, the client's comparison sample has fewer (or more) words than the database samples. Thus, comparison of word-based measures such as number of omissions, number of mazed words, and NDW is not as valid. NDW, in particular, is only valid if NTW is comparable. There are utterance-based measures such as abandoned utterances and pauses within/between utterances which would be more valid if you controlled by the same number of utterances. Other measures are ratios or percentages which are relatively independent of sample length. Although both methods of comparing for sample length are valid, words (NTW) was chosen as the default mostly because of NTW. You can, of course, run the report both ways.
I would like to know why you recommend using total number of words to control for sample length when comparing a client to the database. I have always used number of utterances as the length-control factor (e.g., a 50 utterance sample compared to samples 50 uttererances long in the database). That way, I can use TNW, NDW as points of real comparison between my client and the database. Thx.
If you control by number of utterances and the client's MLU is shorter (or longer) than the database comparison samples, the client's comparison sample has fewer (or more) words than the database samples. Thus, comparison of word-based measures such as number of omissions, number of mazed words, and NDW is not as valid. NDW, in particular, is only valid if NTW is comparable. There are utterance-based measures such as abandoned utterances and pauses within/between utterances which would be more valid if you controlled by the same number of utterances. Other measures are ratios or percentages which are relatively independent of sample length. Although both methods of comparing for sample length are valid, words (NTW) was chosen as the default mostly because of NTW. You can, of course, run the report both ways.