State Library Agencies (StLA) Longitudinal Data File

Abstract

Data from the 50 "state library agencies" plus the District of Columbia Public Library have been collected from the US National Center for Education Statistics (NCES) from 1994-2003. (The DC Public Library also has data in PLDF3 because it has functions like both a state library and an urban public library.) State libraries agencies are commonly referred to as "state libraries" and they perform a variety of functions. They differ from each other in what services they offer a great deal more than other types of libraries and the data are likely to reflect that variety of services and experience.

The data here are from the NCES survey which continues a survey started in 1973 by the Association of Specialized and Cooperative Library Agencies (ASCLA) a division of the American Library Association. Later "editions" were done in cooperation with the Chief Officers of State Library Agencies (COSLA). The various ASCLA/COSLA editions appear to have been published biennially.

There are 510 observations (that is, one library's data for one year) in the dataset with each state and DC Public Library reporting each of the 10 years. A distinguishing characteristic of this dataset, however, is its large number of variables: 647. By comparison, the longitudinal public library datafile, PLDF3, with over 9,000 libraries, has 138,000 observations but only 75 variables.

The 647 includes two variables added during the compilation, state and year, just as PLDF3 and PUSUM, the state public library summary data. All three of these sets of data were issued each year separately and did not have the reporting year included in the data. In order for these data to be used in analyzing trends, the reporting year was added. Each of these three datasets used a different variable for the states--actually there was a total of five such variables for state used in the three datasets so while the original is included, state was created or the compiliations and used in all three. The variable defining the two-character postal code in the StLA data, MAIL_ST, was used to create state for this dataset.

The number of variables, while large, understates the numbers found in the raw datasets. Beginning in 1999, the NCES version of the StLA data include a special kind of variable, a "flag" to indicate that a given variable was imputed. The data in the dataset discussed here, however, have had imputed data removed so that they are as reported by the libraries. "Removing" the imputations meant in this case, if the flags were "R" (as reported by the library, or "E" ("Reported and adjusted by NCES and Census based on edit follow-up") the numbers are in these datasets. If, the flags had any other value, the variable in that instance was set to missing (or "." as it conventionally appears in SAS). After these adjustments, the flags were removed also. In 2003, there were 256 such flags.

Unlike PLDF3 and PUSUM where the universe of collected variables was small, these data changed from year to year, I infer, in response to political winds and of these 643 (less state and year) variables, only 335 were reported each year according to an analysis I did using variable names only for 1994, 1998, 2001, and 2002. (The addition of 2003 now has this file with 337 variables which is impossible and I will have to run this down but time is precious right now. I will get to it as soon as I can.) That is, unlike the longitidinal compilation public library data series which involved close reading of each year's documentation to compare definitions, in this case, only the variable names were used. I have been assured that the folks doing the annual compilations have never reused a variable name in that period inspite of variables' being dropped and others' being added, hence, this strategy seems like a good first approximation. Using data always brings out surprises so this list of variables may be adjusted from time to time as experience warrants.

There is an Excel spreadsheet with information on these variables for 1994, 1998, 2001, and 2002 (but not for 2003), one is the list of all variables in each of the four years. These variables are included in a dataset, stla9403, available here. In addition, stla9403s is also available and it is a smaller version of these data with variables reported, it is hoped, in all of the years.

The variables in each are not given here in the detail given in the other longitudinal files in the interests of speed. It would take months to examine the variables for the two datasets with the same care as the other datasets. However, we can define some of the variables by using the labels that are in the master of the data in the SAS format. The edited output from the SAS procedure Proc Contents is included for each dataset. Note that all but one of the variables in stla9403s have labels that give a description of the variable while many in stla9403 do not. This sketchy documentation is not ideal but it is quick and provides a basis to build on.

Documentation for the two datasets:

Links to the annual NCES documentation and data files are found elsewhere on this site.

The Data

The two sets of data are available in the formats listed below. If you need them in another format, let me know.

February 16, 2005