A***A
发帖数: 98 | 1 Educational measurement is quite different from psychological measurement (in
a strict sense) both in theory and in pratice. The state testing program that
I've worked on has a routine sample size of 30,000 for each form of the test
in each administration. And think about the SAT or tests made at ETS, while a
typical social psychological experiment in JPSP has a sample size that's
roughly 1% of a typical state testing program.
CTT and IRT are complimentary to each other. For practical stuatio | a*e
发帖数: 431 | 2 精彩精彩，多谢多谢。
【在 A***A 的大作中提到】
: Educational measurement is quite different from psychological measurement (in
: a strict sense) both in theory and in pratice. The state testing program that
: I've worked on has a routine sample size of 30,000 for each form of the test
: in each administration. And think about the SAT or tests made at ETS, while a
: typical social psychological experiment in JPSP has a sample size that's
: roughly 1% of a typical state testing program.
: CTT and IRT are complimentary to each other. For practical stuatio
| r******s
发帖数: 2155 | 3 真的专家发话了。very informative.
不过你的例子都是测验方面的。测验和心理测量的区别是前者有正确答案而后者没有。
比较一下智力测验和likert-type scale.例如多项选择，对智力测验来说，如果A是正确
答案，那么其他的就都是错的。于是就可以用IRT考察每个选项的区分度。但是心里测验
在指导语里就明确指出答案没有正确与错误之分，要按照自己的真实感受来填写。所以，
我很外向和我很不外向都可能是正确答案，因为每个人的大案都可能是不同的。
你对alpha的攻击个人认为略有偏颇。方法毕竟只是方法，其功效由目的所左右。
测验的目的是区分高能力者和低能力者，所以每个题目以及选项的区分都很重要。
但心里测量的目标是精确测量，所以题目之间的consistency是必要的。在没有更好
的假设之前，没有理由区别对待每个题目。
所以回到我以前提过的，因为目的不同，所以CTT和IRT的功用不同。不存在一个比
另一个更优越的问题。在测验领域，我同意IRT比CTT更有效，但在信里测量方面，
IRT是不合适的。
(in
that
a
least
starts
a
trait
Instead
【在 A***A 的大作中提到】
: Educational measurement is quite different from psychological measurement (in
: a strict sense) both in theory and in pratice. The state testing program that
: I've worked on has a routine sample size of 30,000 for each form of the test
: in each administration. And think about the SAT or tests made at ETS, while a
: typical social psychological experiment in JPSP has a sample size that's
: roughly 1% of a typical state testing program.
: CTT and IRT are complimentary to each other. For practical stuatio
| A***A
发帖数: 98 | 4 Well, I'm still learning. It's great to find a home on this board to discuss
things that I care about.
确
验
，
Right. The 1st generation of IRT models developed in the 40s and 50s could
only model "right"/"wrong" dichtomies. The 2PL&3PL models, and the normal
ogive model, among others, are perhaps the most familiar form of IRT that
people encounter in educational testing. And as a concequence people tend to
assume that IRT is only for right or wrong responses.
Modern IRT models developed in th | a*e
发帖数: 431 | 5 你们俩都很猛，长见识了。
discuss
。
正
测
以
to
flexible
IRT
models,
also
IRT
estimated
【在 A***A 的大作中提到】
: Well, I'm still learning. It's great to find a home on this board to discuss
: things that I care about.
: 确
: 验
: ，
: Right. The 1st generation of IRT models developed in the 40s and 50s could
: only model "right"/"wrong" dichtomies. The 2PL&3PL models, and the normal
: ogive model, among others, are perhaps the most familiar form of IRT that
: people encounter in educational testing. And as a concequence people tend to
: assume that IRT is only for right or wrong responses.
| r******s
发帖数: 2155 | 6 Great job. I can tell that you are pretty knowledgable about IRT with a
passion. Good for you.
I'll check that article out for sure when I have a chance. In the mean time,
I think there is a lot of IRT 启蒙 needed in this board. You mentioned a
bunch of terms that I have no clue.
I'm quite familar with CTT,but I'm also an open-minded researcher. Your
advertisement does sound appealing and coherent. But that still leaves me the
doubt that if IRT is superior to CTT in all acounts, why it's not domin
【在 A***A 的大作中提到】
: Well, I'm still learning. It's great to find a home on this board to discuss
: things that I care about.
: 确
: 验
: ，
: Right. The 1st generation of IRT models developed in the 40s and 50s could
: only model "right"/"wrong" dichtomies. The 2PL&3PL models, and the normal
: ogive model, among others, are perhaps the most familiar form of IRT that
: people encounter in educational testing. And as a concequence people tend to
: assume that IRT is only for right or wrong responses.
| m******y
发帖数: 526 | 7 Nice piece ah, I like this a lot.
Just add some comments:
1) The difference between CTT and IRT also can be detected from their basic
models/functions: since under CTT, X=T+E. What CTT takes care of is a summary
of observable scores. That is to say, CTT deals with the test as a packet of
items. While under IRT or also the factor analysis, X=lamda*F +U, where F is
the latent trait, so X here refers to one specific/individual item. That is to
say, IRT deals with the test item by item.So we don't h
【在 A***A 的大作中提到】
: Educational measurement is quite different from psychological measurement (in
: a strict sense) both in theory and in pratice. The state testing program that
: I've worked on has a routine sample size of 30,000 for each form of the test
: in each administration. And think about the SAT or tests made at ETS, while a
: typical social psychological experiment in JPSP has a sample size that's
: roughly 1% of a typical state testing program.
: CTT and IRT are complimentary to each other. For practical stuatio
| r******s
发帖数: 2155 | 8 Great! Another expert. I like your point 1 in particular because I'm into
Covariance Structure Modeling a lot.
Can you explain more about the difference between factor analysis and IRT?
summary
to
worked
easier
more
it
tests.
CTT.
test
while
be
IRT
【在 m******y 的大作中提到】
: Nice piece ah, I like this a lot.
: Just add some comments:
: 1) The difference between CTT and IRT also can be detected from their basic
: models/functions: since under CTT, X=T+E. What CTT takes care of is a summary
: of observable scores. That is to say, CTT deals with the test as a packet of
: items. While under IRT or also the factor analysis, X=lamda*F +U, where F is
: the latent trait, so X here refers to one specific/individual item. That is to
: say, IRT deals with the test item by item.So we don't h
| A***A
发帖数: 98 | 9 summary
to
that's right on target. :-) well, no wonder it's called Item response theory,
eh?
worked
easier
more
Despite the fact that 1PL and the Rasch model is functionally equivalent,
Rasch arrived at the logistic item characteristic curve from a fundamentally
different perspective, while Birnbaum's logistic models are derived to
approximate the normal ogive model, especially when it's written explicitly
with that D=1.7 constant to make the logistic cdf and the normal cdf as close
as possible. | A***A
发帖数: 98 | 10
good to know this. most of my time i'd spend on structural equation models.
to push it to the very extreme, unidimentional irt is a factor analysis with 1
factor.
【在 r******s 的大作中提到】
: Great! Another expert. I like your point 1 in particular because I'm into
: Covariance Structure Modeling a lot.
: Can you explain more about the difference between factor analysis and IRT?
:
: summary
: to
: worked
: easier
: more
: it
| r******s
发帖数: 2155 | 11 ^_^
I'd be glad to talk about it. Don't raise your expectation too high though
because I'm more of a informed consumer than a methodologist like you.
Any idea on testing moderating effect in SEM? It's a red hot topic right now
and a much needed one. The common consensus seems still that if you include
interaction term in your equation, use regression instead of SEM.
1
【在 A***A 的大作中提到】
:
: good to know this. most of my time i'd spend on structural equation models.
: to push it to the very extreme, unidimentional irt is a factor analysis with 1
: factor.
|
|
