江郎才劲
请教一个关于信度检验的问题
问卷量表设计是基于某一理论的,比如说我的量表某一构面要调查员工满意度,那么这一层面的问题(题项)可能包括:对工资水平是否满意;对工作环境是否满意;对人际关系是否满意;对成长前景是否满意等。但在这一构面的内部一致性检验的时候发现以上问题中的某一问题(如对工作环境是否满意)导致这一构面的信度较低。那么在正式量表(或本次问卷分析)中是否应该删除这一题项?如果不删除,则信度较低,问卷设计不理想;如果删除,则这一构面反映的内容又不全面,不符合理论要求。
望大家解答。
谢谢!
anita_jiu
[quote]引用第0楼江郎才劲于2006-12-06 17:46发表的“请教一个关于信度检验的问题”:
请教一个关于信度检验的问题
问卷量表设计是基于某一理论的,比如说我的量表某一构面要调查员工满意度,那么这一层面的问题(题项)可能包括:对工资水平是否满意;对工作环境是否满意;对人际关系是否满意;对成长前景是否满意等。但在这一构面的内部一致性检验的时候发现以上问题中的某一问题(如对工作环境是否满意)导致这一构面的信度较低。那么在正式量表(或本次问卷分析)中是否应该删除这一题项?如果不删除,则信度较低,问卷设计不理想;如果删除,则这一构面反映的内容又不全面,不符合理论要求。
望大家解答。
.......[/quote]
What test have you used for internal reliability, please? By the way, how low is your result?
江郎才劲
Crobach Alpha, 在没有删除题项时约为0.4,如果删除题项能达到0.6
anita_jiu
[quote]引用第0楼江郎才劲于2006-12-06 17:46发表的“请教一个关于信度检验的问题”:
请教一个关于信度检验的问题
问卷量表设计是基于某一理论的,比如说我的量表某一构面要调查员工满意度,那么这一层面的问题(题项)可能包括:对工资水平是否满意;对工作环境是否满意;对人际关系是否满意;对成长前景是否满意等。但在这一构面的内部一致性检验的时候发现以上问题中的某一问题(如对工作环境是否满意)导致这一构面的信度较低。那么在正式量表(或本次问卷分析)中是否应该删除这一题项?如果不删除,则信度较低,问卷设计不理想;如果删除,则这一构面反映的内容又不全面,不符合理论要求。
望大家解答。
.......[/quote]
你的问题里有好几个问题。Employee job satisfaction can involve different dimensions, as you said above, salary, physical working environment, personal relationship with others, prospective development, etc. One question emerges here: do you break down these dimensions into multiple items? For example, when you measured salary, did you just ask your respondents whether they were satisfied with their monthly salary but not anything else? If you used multiple items to measure one dimension, then the number of items certainly has impact on your internal reliability test. When your items are less than 10, it is not surprising that you have small Alpha value.
If you had not used multiple items, but simply measured job satisfaction by the dimensions you mentioned above, then now you need to think objectively about the results of reliability test. Numbers are numbers, they don't lie. But the more important issue here is how the researcher interpretes his/her results. Inadequate interpretation certainly can mislead readers; what is worse is that inadequate interpretation very often masks the true story behind the scenes.
The common practice of Alpha reliability value is above .7. But for exploratory research, this could be too restricted and therefore lower value is more practical. There are various possible explanations of low Alpha value. The item you tested is seen differently by your respondents from other items. The number patterns are different. It could be that:
1. the scale/question you examined is not reliable due to inconsistent wording among your questions, or
2. the scale/question you applied is reliable in one population but not valid in your case.
You could indicate that future research needs to design a 'better' scale in a more consistent manner. Alternatively, the scale you used in your research needs to be validate in other population. I would also like to say that if the scale shows a different variance pattern to others, why couldn't you delete this scale and report a higher and more stable scale?
abel
尝试分解为多个主题吧。如果你了解常用的信度是如何计算的,也许你会知道怎么处理这种问题,一般删除题目可以提高信度的应该坚决不要或者替换,因为也许这个题项测试的和主题太不相关了,甚至测量的是另外的问题。
如果一个完整的量表,信度只有0.6的话,建议还是做大面积的修正好。
hgsean
我也遇到过同样的问题。我是学英语教育的。最近在做动机问卷的时候发现,成绩动机层面的某个题目,在成绩动机单独做克隆巴赫系数检验的时候,删除某个题目可以使该维度的信度增加,但是如果测算整个问卷的克隆巴赫系数,这个题目却没有问题。也想请教各位。
hgsean@126.com
domakelove
有道理!
关联度很重要
不是说一个常用的量表就可以直接拿来用的
同样一个问题或是一个量表中的变量
放到其他地方可能不合适
所以 还是要根据调研目的来确定问题
[quote]引用第3楼anita_jiu于2006-12-07 18:02发表的“”:
你的问题里有好几个问题。Employee job satisfaction can involve different dimensions, as you said above, salary, physical working environment, personal relationship with others, prospective development, etc. One question emerges here: do you break down these dimensions into multiple items? For example, when you measured salary, did you just ask your respondents whether they were satisfied with their monthly salary but not anything else? If you used multiple items to measure one dimension, then the number of items certainly has impact on your internal reliability test. When your items are less than 10, it is not surprising that you have small Alpha value.
.......[/quote]