请教：Keywords结果阐释

[Pennylin]

用WordSmith的Keywords产出如下结果：

[FONT=宋体]N [/FONT][FONT=宋体]WORD [/FONT][FONT=宋体]FREQ. [/FONT][FONT=宋体]CH53BA~1.TXT % [/FONT][FONT=宋体]FREQ.[/FONT][FONT=宋体]NS50B0~1.TXT % [/FONT][FONT=宋体]KEYNESS [/FONT][FONT=宋体]P[/FONT]
[FONT=宋体]1 [/FONT][FONT=宋体]WOMAN IN THE PICTURE [/FONT][FONT=宋体]13 [/FONT][FONT=宋体]0.02 [/FONT][FONT=宋体]0 [/FONT][FONT=宋体]24.4 [/FONT][FONT=宋体]0.000001[/FONT]
[FONT=宋体]2 [/FONT][FONT=宋体]DRAW A PICTURE OF [/FONT][FONT=宋体]13 [/FONT][FONT=宋体]0.02 [/FONT][FONT=宋体]0 [/FONT][FONT=宋体]24.4 [/FONT][FONT=宋体]0.000001[/FONT]

其中，Keyness值和p值多少为具有（显著）差异？为什么有些p值为0？

多谢指点！

 

[xujiajin]

回复: 请教：Keywords结果阐释

进入词表的（即你看到的）都是显著的。

keyness （chi-square or LL）的临界值：
“chi-square 值越大越好，说明所比较的两个项目差异越大（具体有关原理，如果有兴趣的话，可参看相关统计书，如有关observed frequency和expexted frequency的问题）。

不过两个项目存在显著差异的前提是，该卡方值至少必须大于临界值3.84（比如卡方值为5，当然越大越好）。即在p = 0.05的显著性水平上，两者具有显著差异；如果能大于6.63或10.83则显著性更高。

* 5% level; p = 0.05; critical value = 3.84
* 1% level; p = 0.01; critical value = 6.63
* 0.1% level; p = 0.001; critical value = 10.83”

http://www.corpus4u.org/showthread.php?t=4213

p值为零是因为保留小数点位数的缘故。

 

[Pennylin]

回复: 请教：Keywords结果阐释

谢谢指点！

再请教：
在Keywords的setting里，可以选择chi-square或log likelihood值。请问值有什么区别？

 

[xujiajin]

回复: 请教：Keywords结果阐释

Both log likelihood and chi square tests will do for keywords generation purpose. Chi-squared test is a classic and 'robust' test for significance of difference, but log likelihood is thought to be able to yield improved results for significant difference test. Honestly, lay people like us don't see much difference of the two. Regarding the keywords generated, there are virtually the same, esp. for the top ones. However, in WS and AntConc, LL is the preferred/default setting. That is to say, you don't have to bother about the setting, just let it be like that.

If you are desperately interested to know the difference, you can read the following book chapter with my annotations.

 

[Pennylin]

回复: 请教：Keywords结果阐释

Thanks for being so helpful and patient in answering all those questions, Dr Xu.