无肉不欢
本人硕士论文研究题目之一是: 中韩两国大学生的博客使用动机有何差异?
调查问卷中共设了22个动机项目,在使用因子分析进行成分输出时,出现了不应该是一类的项目被归在同一个因子成分下的情况,如:接触性因素下出现了“为了保存照片”这个项目
为了向朋友传递我的消息 .766
为了了解朋友的近况 .752
为了保存照片 .705
乐于看到访问者的回应 .678
便于与熟识的人沟通 .626
18号论文就要预审了,着急的很,请各位高人指路!小女子这厢有礼了~
dinglin66
讲得不是很清楚,在说详细点~~
无肉不欢
Rotated Component Matrix(a)
Component
1 2 3 4 5 6
向朋友传递我的消息 .766 .008 -.092 .189 -.022 -.054
了解朋友的情况 .752 .063 -.086 .086 .099 .036
保存照片 .705 .097 .241 -.125 -.008 .078
乐于看到访问者的回应 .678 -.104 .157 .193 .014 -.008
方便与熟识的人交流 .626 -.074 .138 .196 .093 -.040
赶时髦 .073 .891 .059 .041 .042 .080
别人都开博客 .055 .863 .056 .033 .102 .059
如果没有别人觉得奇怪 -.203 .663 .060 .030 .132 -.210
为了获得多样信息 .067 .019 .821 .135 .035 .130
资源共享 .182 .092 .802 .108 -.010 -.177
为了读好的文字欣赏图片.337 .026 .579 .107 -.118 .336
习惯性的 -.161 .067 .526 .101 .328 -.039
主人公的感觉 .117 .001 .170 .767 .132 .187
表现自我个性 .002 .447 .094 .651 -.101 .004
拥有属于自己的空间 .408 -.267 .191 .581 -.061 .013
展示另一个自我 .394 .004 .103 .567 .175 .045
逃避现实 -.163 .138 .120 -.077 .788 -.051
减少孤独感 .328 .133 -.002 .128 .705 .083
不好开口的事情可以倾诉.384 .060 -.029 .399 .552 .181
抒发个人情感 -.152 -.254 -.118 .094 -.087 .708
调节心情休息 .088 .166 .211 .182 .106 .707
打发时间 .067 .521 -.035 -.119 .226 .532
Extraction Method: Principal Component Analysis.
Rotation Method: Varimax with Kaiser Normalization.
a Rotation converged in 7 iterations.
无肉不欢
这是输出的结果, 我觉得“保存照片”不应该与另外4个归在同一个要因里面阿。。
netcow
这种情况现实当中很常见。可能是由于样本含量不够或者自行设计问卷的调查项目的问题。
yihui
保存照片的目的大多不也是让别人看么,所以也可以算作“接触性因素”啊
无肉不欢
谢同学说的有道理啊, 我的研究题目具体是这样的:
1.中韩两国大学生的个人主义-集体主义倾向如何? 两国是否存在差异? 我用了分散分析,平均分析和T-test~
2.中韩两国大学生的博客使用动机和活动如何?两国存在什么差异?
在这里我是应该分为韩中两个集团,单独进行因子分析进行比较呢? 还是整体进行因子分析,然后用T-
test?
3 中国两国大学生的个人-集体主义倾向的差异与博客使用动机和活动的差异有什么联系?
我想用相关分析,不知道对不对?
哎~都要疯了~
anita_jiu
[quote]引用第6楼无肉不欢于2006-11-01 16:13发表的“”:
谢同学说的有道理啊, 我的研究题目具体是这样的:
1.中韩两国大学生的个人主义-集体主义倾向如何? 两国是否存在差异? 我用了分散分析,平均分析和T-test~
2.中韩两国大学生的博客使用动机和活动如何?两国存在什么差异?
在这里我是应该分为韩中两个集团,单独进行因子分析进行比较呢? 还是整体进行因子分析,然后用T-
.......[/quote]
首先,想说明一下,我也在学用因子分析,不过不是主成份分析,而是`真的`因子分析 - Principal Axis Factoring。There are differences between Principal compoennt analysis and PAF, conceptually and mathematically. 乐意将我观点与你探讨,大家一齐讨论。
你所讲的,在一个component里,出现了一个`不应出现`的variable,这有好几种可能性。建议你先examine the original correlation matrix and check whether this variable (=`不应出现`的) has high correlations with the other variables shown in this component. I suspect the answer would be positive. Then the next question is whether this variable has any conceptual linkage/relationship with other variables. As YiHui suggested, in your component, many variables share the common variance of 'being touched with', e.g. send messages to friends, keep in contact with friends, etc. Perhaps for your respondents, 'saving photos' may have further meanings (as what Yihui suggested), e.g. being shared with friends or show friends what they are up to, or pleasure, funny moment/experience etc. Therefore, it is not surprising that this variable is extracted and rotated in this component.
I'm not sure whether you would like to further investigate your data. Have you tried to split the whole sample into two sub-groups, i.e. Chinese vs. Korean, and try to see whether PCA would produce two different results (i.e. different component structure)? If the components are different between the Chinese and Korean, what are the potential explanations? If the components are not massively different, why? E.g. people from these two countries broadly share common culture values? Or this commonality is due to your sample? By the way, what is the nature of your sample? E.g. sample sizes in these two groups? Are they equally split? If not, which group has more respondents? This group could possibly dominate your analysis results. And this is why I suggest you split your whole sample into two small groups.
Just some thoughts about your research. Not sure whether they would be useful or not. Good luck...By the way, have you named/labelled your components yet?? What are you going to do after PCA - more specifically, what conclusion would you wanna to draw from the analysis?
无肉不欢
十分感谢 ANITA-JIU~
我尝试了把所有的数据放在一起分析(共304), 还有分为韩国中国两类分析(152,145) 所得的结果都不一样啊.
反正现在是一团糟,哎~
还有个问题是: 在 Communalitiesli结果表中当 extraction低于0.4的时候,这个项是不是就可以排除了?
0.4的标准是研究者自己决定的吗? 是不是也可以根据自己的情况定成其他标准呢?
anita_jiu
对不起,晚了回复,今天就邀去好朋友的毕业礼了!
当你的sample是由两个groups组成时,在运行PCA or FA后,得出不同的结果(我指component/factor structure),这是正常的,也可以解释。不要烦,要定下来想为什么结果会有所不同,conceptually。其实即便结果是一样的,也没有什么。结果就是结果,关键的是你怎样正确演译结果。当然,这个大前提是你的数据合符使用这个统计方法。要科学,要公正,(尽量)不要主观。
在你的例子里,有不同(或没有不同)都是好事啊,我认为。你可以从两个不同组别的characteristics(e.g. age, education, blog experience, income etc)里尝试找答案。另外一个是,你的问卷里有没有其他研究的变量?看一看,或许可以找到答案。还有,即使找不到答案,也不用担心。这是`自然调查现象`,你可以将其suggest as a future research direction。如,通过confirmatory factory analysis or simply apply the same research instrucments onto another population.
Not sure whether the above would be useful or not. Good luck...
anita_jiu
另,你提出communality的问题:
我好像没有看到书上有说低于0。4时应该可以将这个量排除。可能是我不仔细吧?!communality的值应该是越高越好。但,当你的sample size大时(如你现在有300人),这个值不高有几种可能性。一个最大的可能性是,这个量本身就可能没有同其他变量产生(相对)高的correlation coefficient (or in other words, this variable doesn't share a high degree of common variance with other variables)。于是经过extraction后,这个值(the communality value)也不高。I suggest you examine this variable's coefficient value in the original correlation coefficient matrix and check whether it has any coefficents greater than 0.3. (a correlation between 0.3-0.4 presents a medium association, however, since your sample is relatively large, this may present a large association).
Factor analysis is a statistical technique broadly based on a researcher's 'intuitive' decision. However, in order to make your results/reserach sound, your 'intuitive' decision should be developed upon logics, if this makes sense to you. This means that yes, on the one hand, you may decide what to do with your variables (e.g. delete x or y), but you have to have a sound reason to argue why you delete x but not y. This is again about balance in your data...
无肉不欢
要哭了~ 不过我要坚强, 下周教授就回来了,到时候应该有救了, 再次感谢 JIU
sober
如果你是做心理学的,那我就要说,你学歪了。心理学中运用统计是用来为你的研究服务的,而不是你的研究要围绕着你的数据转。如果出现你那种情况,不是统计的原因,而是你的项目的原因。你那么问就是本末倒置了。
anita_jiu
[quote]引用第11楼无肉不欢于2006-11-03 15:57发表的“”:
要哭了~ 不过我要坚强, 下周教授就回来了,到时候应该有救了, 再次感谢 JIU[/quote]
你不用担心,我完全明白你的心情。也不用哭,这是学习的一个过程。当我们做调查的`资历`尚浅时,越到你这样的例子,不知道何去何从,这是可以理解的。这也不是想歪,只是可能分析结果跟自己预期的有不同时,以为是negative.
我想说的是,我也刚刚遇到过这样的'negative' outcomes。我用引号,因为我现在觉得其实是unexpected outcomes with personal negative feelings. 这没有什么的。最少,你要思考为什么这个variable会跟其他你觉得不应相关的variables,被extracted to one component. Would this outcome be due to the sample you have? Or the variable is measuring something else etc. ? This is interesting, isn't it? You may at least address these questions, together with some possible explanations. You may possibly suggest the future research direction. If your discussion is developed upon a solid basis of logical arguements, your research would look interesting, wouldn't it? If the question is positive, then some sort of credits might be given to your piece of work.
I'm reading Gorsuch's 'Factor analysis' at the moment. Will come back to you if I find anything interesting...
Be positive...
anita_jiu
Sorry, I'm going to bore you guys here.
Gorsuch states, "The most frequent situation where principal factor solutions are used is exploratory factor analysis. The purpose is to examine the structure of a given domain as represented by the smaple of variables. The long-term interest is to identify basic conceptual dimensions that can be examined in future research." (1983, p121)
See, this is why I said to you in the 9th reply above, as long as you can develop a conceptual understanding of your components - in your context, the next stage is to name your components meaningfully. The outcomes from PCA should be seen as positive. Well then of course, the premise is that your data has met the criteria of factory analysis. This is important. I would suppose you had checked the adequatecy of your data in applying PCA .
无肉不欢
[quote]引用第12楼sober于2006-11-03 16:01发表的“”:
如果你是做心理学的,那我就要说,你学歪了。心理学中运用统计是用来为你的研究服务的,而不是你的研究要围绕着你的数据转。如果出现你那种情况,不是统计的原因,而是你的项目的原因。你那么问就是本末倒置了。[/quote]
我是做传播的, 项目是改不了拉,没有时间了, 我就是在想办法能让得出的数据支持我的论点而已.
无肉不欢
JIU ~ 其实你说的我想过了, 就是给那些分在一起的因子好好综合起来起个合理的名字就好了, 我正在研究中~ THANK U
hwtatm
统计是一种方法论科学,数据一定,方法一定,那么结果也就定了。所以我建议你从专业的角度出发,看看指标的选取和算法有没有问题。注意注意指标。
江郎才劲
[quote]引用第6楼无肉不欢于2006-11-01 16:13发表的“”:
谢同学说的有道理啊, 我的研究题目具体是这样的:
1.中韩两国大学生的个人主义-集体主义倾向如何? 两国是否存在差异? 我用了分散分析,平均分析和T-test~
2.中韩两国大学生的博客使用动机和活动如何?两国存在什么差异?
在这里我是应该分为韩中两个集团,单独进行因子分析进行比较呢? 还是整体进行因子分析,然后用T-
.......[/quote]
请问第二个问题是怎么解决的?我也想知道