hgsean
各位大侠
大家好!
本人是学英语教育的。对于统计知识知道皮毛,但是由于做项目需要做统计,所以没有办法对于我这个因为要逃避学数学选择英语专业(当然更主要的是喜欢英语专业)。呵呵
最近和一个朋友合作做一个关于英语学习动机的问卷,我们进行了因子分析,但是因子分析结果并不是象我们事先想象的那样,在一个维度下面的题目,通过探索性因子分析正交和斜交都归到这个维度下面。有的题目跑到另外维度里面了。可能是这些题目之间的相关性强的原因?所以我们特别想知道是不是由于统计的数据的原因?当然措辞很重要。但是如果我加入一个:你对现在的涨工资之都满意么?,这个和英语学习动机显然没有联系,但是会不会在问卷统计的时候这个题目也归入到一个动机中呢?我们用的是6点Likert等级制。
不知道我是否叙述清楚了。
举例: Component
Component
1 2 3
AN18 .891 .
AN16 .861
AN14 .830
AN19 .506
AN11 .817
AN12 .796
AN10 .654
AN13 . .576
AN31 .608
AN30 .563
AN26 . .529
AN9 .512
AN15 .485
例如这是我们银子分析的一部分成果,但是an15应该是第一个因子中的题目,但是现在却到了第三因子;为什么呢?假设我们的措辞没有问题,那么是是什么原因呢?
如果我加入题目;我对现在涨工资很满意。6非常同意 5同意 4有点同意 3不太同意 2 不同意 1非常不同意
那么这个题目是不是一定是因子分析后最后那个维度也不能包含的题目?被甩出来需要删除呢?
谢谢了!
我的邮件hgsean@126.com
anita_jiu
1. It is delighted to witness that statistics are widely applied in various fields.
2. Theory Doubts - I think we shall be 'braver' when facing 'theory doubts'. In social science research espeically that human behaviour is involved, it is often that we find different results from those illustrated in well-established theory. One possible, and perhaps main, reason is that human behaviour is full of complexity, varying with contexts; you may experience this complexity through the various dimensions developed in the theory you are interested in.
However, this difference shouldn't be seen as a neagtive sign to theories, indeed. More importantly, it should direct us to the next question - why, why in your context/research, results are different from the theory. Questions should be developed, e.g. is it because of a different sample you used, what is the basis of your theory when it was developed etc.? A thorough investigation should help us better understand the theory under examination.
3. Winthin your components, variables that are extracted and rotated in a particular factor are not from the same dimension - try to find any hidden relationship between the variables within one factor, e.g. check your original correlation matrix, see the variables mostly relate to AN15. They would be those illstruated in Component 3. Try to see the original measures, think why they are correlated.
anita_jiu
There are a few more questions here in deciding whether or not to delete 'AN15':
1. check the original correlation matrix, see how well 'AN 15' is correlated with other variables. What is the R² value of 'AN15'? How 'big' is it?
2. check the communality value of 'AN15'.
3. what is your component loading value? >.4 or ...?
4. How big is your sample?
5. any component structure difference if you delete 'AN15' (if you decide to delete 'AN15' based on sound criteria?