home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!SAKURA.CC.TSUKUBA.AC.JP!JDS
- Content-Type: text
- Content-Length: 2405
- Message-ID: <9208280847.AA18651@sakura.cc.tsukuba.ac.jp>
- Newsgroups: bit.listserv.stat-l
- Date: Fri, 28 Aug 1992 17:47:13 JST
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: "john.derrick.shillaw" <jds@SAKURA.CC.TSUKUBA.AC.JP>
- Subject: advice sought on language test analysis
- Lines: 50
-
- [I posted this 3 days ago but I have a feeling that it didn't get
- through to the list. My apologies if it did.]
-
- I'm working on research into the effectiveness of the different English
- language tests that we use in the Center where I work and although I
- have some experience in statistics, I have three problems where I feel a
- little out of my depth. I'd be very grateful for advice on the
- following questions. For analysis, I have access to SPSS/PC 3.0 and
- Testfact (Scientific Software).
-
- 1. I have 4 parallel tests of listening, each administered to a sample
- of the total population, each sample balanced for ability. Each test is
- made up of 3 sub-tests identical in format and length. I have
- correlated the 3 sub-tests on each test and corrected for attenuation.
- I now want to get the average correlation from the sub-tests (Form A,
- part 1 + Form B, part 1 ...) on the 4 tests to get some idea of the
- relationship between the sub-tests for the total population. Is this
- valid or is some further correction necessary?
-
- 2. I want to find out if I can reduce the 3 sub-tests to 2 in the final
- form and so need to discover which of the sub-tests are the best
- predictors of performance on the test. To this end, a) is it valid to
- regress the 3 sub-tests on the test total, and if so, b) how can I get
- around the problem that part 2 of the test is only half the number of
- items of the other two parts and its contribution to the total will
- therefore automatically be reduced? and c) more generally, is there a
- way of minimizing the relationship between the dependent and independent
- variables? If regression is out, is there another way I can achieve the
- same objective?
-
- 3. If I want to test for unidimensionality of trait on the listening
- tests, should a factor analysis be done on the inter-item tetrachoric
- correlation matrix, or on the total scores on the 3 sub-tests?
-
- Could somebody please tell me the BITNET address for the EDSTAT-L
- listserver. I subscribed some months back, then set the mail off for
- the vacation and now I've lost the address!
-
- Thanks in advance for any help.
-
- ======================================================================
- John Shillaw
- Associate Professor
- Foreign Language Center
- University of Tsukuba
- Ibaraki-ken
- Japan
- e-mail: jds@sakura.cc.tsukuba.ac.jp
- fax: +81-(0)298-53-6616
- ======================================================================
-