Application Series 12


EDUCATED IN THE CASELESS CASE

Niels Veldhuijzen, Statistician at Cito, Arnhem, The Netherlands, uses SIR for the National Assessment Project. He explains why he prefers the caseless, or relational, SIR structure to the traditional hierarchical or case structure.

What is Cito?

The National Institute for Educational Measurement in the Netherlands (better known under its Dutch acronym CITO) is a semi-governmental institute which provides the Dutch educational system with many varieties of tests and other evaluation instruments. These measure abilities of individual pupils and monitor school performance. More than 300 people work at Cito; 6% of them at the Measurement and Research Department as statisticians and psychometricians. Cito works for all kinds and types of schools except universities.

What is the National Assessment Project?

The National Assessment Project is a cyclic project performed on behalf of the Dutch Department of Education. Every five years, a survey is held among eight-year-old and twelve-year-old primary school children. Its main purpose is to measure the output of the Dutch educational system with regard to subjects like language, arithmetic, geography and physical training. A secondary purpose is to compare results over time, in order to see if there are shifts in the general output levels for the two populations.

What does the National Assessment Project?

The project has two main stages: instrument construction and measuring. Instrument construction is the process of producing test items (questions with direct answers and tasks such as essays which are rated) which are meant to elicit responses from pupils, followed by statistical and psychometric analyses in order to turn the items into measuring instruments with known characteristics. The second stage consists of taking the measuring instruments to the pupils, gathering the responses and analyzing and interpreting these responses.

About the data

In each survey, between four thousand and ten thousand pupils take part. For each subject many items are available. As it is impossible to confront all pupils with all items, an incomplete design is used. Items are divided into several subsets, and different subsets are given to different subsets of pupils. The use of sophisticated psychometric models (so-called item response models) makes it possible to link responses on one subset of items to those on other subsets of items pertaining to the same subject matter. Due to the incompleteness of the design and the large numbers of items and pupils, the structure of the data gathered is very complex.

A Case for a Caseless Database

The National Assessment databases have evolved over many years of experience and development, and one of the interesting features of today's databases is that they are caseless, or relational. From the two stages of the National Assessment Project, instrument construction and measuring, it can be deduced that there are two distinct kinds of objects in the project, the test items and the pupils. These two objects are linked, of course, by a third object: the responses of the pupils to the items. A caseless database allows all three objects to be incorporated into a single database with very small keys, or links, between them. It also allows for the incorporation of metadata, detailed information about the objects.

Responses

Of paramount interest for the project are the responses the pupils gave to the test items. Two very simple but very large record types have been defined; one to hold responses to specific questions and one to hold ratings of a task. Because items are divided into subsets, each record contains at least two key fields: a pupil id and a subset id. The ratings record type has the rater id as the third key field. By means of SIR's full report generator, it is very easy to collect data pertaining to one pupil in an external file. The key fields provide the links to the other two objects I mentioned earlier: pupils and items.

Pupils and Schools

One record type has been defined to hold background information about each pupil such as sex, age, ethnicity. As some school characteristics are also studied, an important background variable is the school id, which is incorporated in the pupil id. The school id connects pupil records with a school record type, which holds school characteristics such as the text books used by the school (CAT VARS does splendidly here!) and a school category.

Subsets and items

A record type has been defined for item subsets with the subset id as its only key field. It contains item ids for this subset. The order of the item ids corresponds with the order of the items in the printed test booklets for this subset. Because of the incompleteness of the design, some items are part of many subsets. These item ids point to item records, storing data such as the kind of item (multiple choice, open ended question), the code of the correct response for a multiple choice answer, and the possible response categories for other items.

Data extraction

Thanks to the caseless, modular structure of the databases, data can be combined in many ways. SIR's full report generator is the indispensable tool here. Collating records into lines to be put in external files, aggregating data over variables, and many other data manipulations are easy to do, resulting in data files with any format my colleagues and I like.

This is especially important for the psychometric analyses of test items, which at Cito is mostly done by computer programs developed in house and require various and complex files in a special format. A suite of PQL retrievals and programs has been written that makes construction of these files very easy.

In fact, a two-stage procedure is followed. In the first stage, the data and metadata needed are specified with the help of PQL-menus. The information supplied in the menus is used in a procedure generator, which writes the PQL retrievals and programs to be used for data extraction. In the second stage these generated procedures are run in batch mode, mostly at night, to produce the required files. The next day, all the files which we have specified for the execution of our psychometric analyses are waiting for us!

Updating the databases

Per subject matter and per age group, a survey is held every five years. The new data are added to the databases, which will grow over the years. Adding data goes smoothly with the help of the ADD REC utility. The data files to be added to the databases are prepared by data entry specialists with the help of a dedicated SIR input database which provides input screens with FIELD INPUTs and immediate data validation.

Final Remarks

The National Assessment Project databases are installed on a personal computer with 80386 and 80387 processors, a 33 MHz clock and a fast 500Mb SCSI-drive. At the start of the project, my PC was called 'the BMW of PCs'; now it is called utterly out of date. But thanks to the modular structure of the databases, the small key fields, the use of program generators and, most of all, thanks to the many features of SIR and its beautiful PQL, managing the National Assessment databases with SIR is a pleasant job.

For more information on The National Assessment Project contact:

Niels Veldhuijzen,
Cito, Measurement & Research Dept
PO Box 1034
NL-6801 MG Arnhem
The Netherlands

Back Sir Home