怎样使用SPSS

[kavona]

统计出两个语料各自使用被动语态的数量后， 怎样比较其有否差异，是否能用T-TEST，怎样用？ 进行比较的两个数据是否都要计算成百分比

[清风出袖]

应该用卡方检验吧？你只有两个样本么？！

[xiaoz]

Chi squared or log-likelihood tests will do. If you do not know how to use SPSS, you can try the LL caculator below:
http://ucrel.lancs.ac.uk/llwizard.html

For "Frequency of word", enter the number of passives in your two corpora, and for corpus size, enter the number of words in the two corpora.

If you get a score above 3.84, the difference is statistically significant at significance level 0.05, if the score is greater than 6.64, the sinificance level is 0.01, if the score is above 10.83, the sig. level is 0.001.

[kavona]

thank you very much

[kavona]

is sig. 0.05 the significance level of SPSS? by the way, most of the score i got is lower than 3.84, does that mean there is no significant difference between the two coupus?

[xiaoz]

SPSS is the name of a statistics software package. significance level 0.05 means you can be 95% confident (should be at least 95%) that the difference is statistically significant, i.e. the difference is not caused by chance.

Here is an example. In the Chinese Learner English Corpus (CLEC), which contains 1,070,602 words, there are 9,711 instances of passives; in the Louvain Corpus of Native English Essays (LOCNESS), which contains 324,304 words, there are 5,465 instances of passives. Is the difference statistically significant? If you compute the log-likelihood score, it will be 1235.6 (for 1 degree of freedom - if you are using UCREL LL calculator, the d.f. is 1), much greater than 3.84, or even 10.83 for significance at p<0.001. So now you are over 99.99% confident the difference is highly significant statistically. If you bring the raw frequencies of passive in the two corpora to occurrences per 100,000 words, you will find that there are about 907 instances of passive in CLEC but 1,685 instances in LOCNESS. So you can say that passives considerably under-used by Chinese learners of English.

[kavona]

excuse for my slowness in computing. but i got a score of 0.52, much lower than 3.84, dose this mean there is no great difference ? thank you?

[xiaoz]

Then, the difference is not significant.

[kavona]

is the order of coupus 1 and corpus 2 matter? it seems that the overused and underused is depended on the order of which number input first. shall i fix that order or choose one corpus as a standard one?

[xiaoz]

The order is not an issue. But the data types (word counts and frequencies of passives for example) must match each other.

[asan82]

一个变量两组--独立样本T检验
一组两个变量--成对样本T检验
一个变量下三组或三组以上--方差分析
至于多元的,那就用卡方检验
我们老师上课总结的,不知道我理解得对不对
(可能把复杂的问题简单化了)

[kavona]

thank you !