State Summary/State Characteristics Longitudinal Data File (PUSUM)

Robert E. Molyneux

Abstract

This page documents a longitudinal set of state summary public library data constructed from two sources. One is the annual Public Library Survey formerly collected by the National Center for Education Statistics (NCES) under the Federal State Cooperative System (FSCS) program and more recently from the continuing series published by the Institute of Museum and Library services through the Public Libraries Survey. This compilation is of the State Summary/State Characteristics Data File, that is Public Use Summary (hence, I conclude, PUSUM which convention NCES and IMLS have named files in this series). The dataset now contains 1,560 observations (an observation being summary data from the libraries in one state for one year), with 85 variables per observation, although not all variables were collected all years. These data are from the states, the District of Columbia, and occasionally outlying territories.

The summary data were originally compiled starting in Fiscal Year (FY) 1987. For reasons discussed shortly, the set of data discussed here only have the data from FY 1992-FY 2020 or 29 years. The data are available in formats discussed at the bottom of the page at Distribution and Formats.

The state summary data are fundamentally different from the Public Library Data File which is a compilation of data from over 10,000 public libraries collected from FY 1987-FY 2020. While PLDF3 is designed to have only unimputed data from all the public libraries, PUSUM includes imputed data.

These summary data are the result of analyzing aspects of those data and other data. Given that not all libraries respond to all survey items, NCES and IMLS have used imputation techniques to compensate for those missing data. Documentation is carried in the various sources cited in the list of NCES and IMLS Public Library Survey documentation elsewhere on this site. PUSUM contains imputed data given that it is attempting to compensate for the missing values for purposes of arriving at national assessments of the state of public libraries.

Knowing these facts, could you replicate these summary data using the Public Library Data File either in its unimputed or imputed forms? Probably not. The documentation on how the imputations were done is improving but does involve data that are masked in the Public Library Data file for privacy reasons. NCES and IMLS have published “unrestricted” and “restricted”data. Data are suppressed in the restricted data. For example, in smaller libraries salary information when the staff is sufficiently small to cause a privacy concern. All the PLDF3 data are from the restricted data.

IMLS has access to data suppressed for individual libraries. Data are suppressed in smaller libraries, for instance, salary information when the staff is sufficiently small to cause a privacy concern. Of course, you could do your own imputations. More telling, though, is the fact that these summary files are also constructed using the Restricted Use data, in part. These Restricted Use data are not available for public use. The great value of PUSUM is assessing the state of public libraries in the U.S. on national and state-levels. Without imputations, there are very few data in this series so I compile them with the imputations. For a universe survey attempting to give the state of the nation's libraries, this is the series to use.

A further complication follows from this practice of imputation. The documentation for years (here 2018) has specifically warned in the much same language:

Comparisons to estimates prior to FY 1992 should be made with caution, as earlier estimates do not include imputations for nonresponse and the percentage of libraries responding to a given item varied widely among states. (p. 6)

According to the documentation from earlier years, there are supposed to be summary data from FY 1987-FY 1991. 1991 is made up of a summary from the reporting libraries so that comparing FY 1991 with FY 1992 is likely to show discontinuities that are an artifact of the imputations themselves and not from any change in the characteristics of those libraries. For FY 1987-FY 1989, in spite of the documentation's saying there are summary data, they are sketchy. FY 1987 has a few variables reported for the 19 states' summaries. Summary data for FY 1988 were not with the rest of the FY 1988 data and I don't know where they are. FY 1989 does have summary data in a separate file as does FY 1990. All imputations could be taken out of the post-1992 data but experimenting with that idea resulted in derivative datasets with very few values. So, we get some data imputed by methods that are becoming clearer but still not unambiguous or we get vanishingly few data.

Variables

The tables below list the variables that were dealt with in constructing PUSUM. The intent was to group all like variables in the various years in one variable name. Rarely was there any ambiguity in reading the documentation from year to year.

Two variables have been added to each year's data in order to ease processing: state and year. (NCES/IMLS variables are in upper case while any non-NCES/IMLS variables are in italics. year was hard coded in one of the programs and is the nominal year of reporting and the same as the fiscal year that appears on publications. The NCES/IMLS library data have been issued as annual publications without explicit acknowledgement in the data of previous years or years that might follow although an infrastructure which allows this practice has improved steadily over the years. For PUSUM, creation of a longitudinal file is relatively straightforward. State is the two-digit postal code. This variable was made from various variables in the dataset as noted below. For recent years, it has been derived from STABR.

A spreadsheet in three formats (pusumvariables.ods, pusumvariables.xlsx, and pusumvariables.csv) presents these variables in a more portable format for the duration of the series.

The following tables have details on the variables in PUSUM. The convention followed is to use the variable name from the current documentation except where the data are no longer collected and the short definitions and notes come, largely, from the current documentation. The variable names have been static for some years.. The wording differs on occasion for variables documented in the earlier years but the definitions overall are pretty consistent. Imputation flags listed in the documentation in Appendix G are not included in this list nor in PUSUM.

As a general rule, the improvements over time in this series and PLDF3 are notable and commendable but, of course, they started from a high level. Both of these series have excellent documentation so anyone using them owes thanks to many people.

Distribution and Formats

January 7, 2023