Modelling Use at Individual Service Points

Citation

Hannabuss, S. (1999), "Modelling Use at Individual Service Points", Library Review, Vol. 48 No. 1, pp. 55-56. https://doi.org/10.1108/lr.1999.48.1.55.13

With the increasing emphasis on performance assessment and benchmarking, it is no surprise to see more data analysis. This report sets out to model use at public library service points. It draws on datasets about branch libraries provided by Birmingham and Bromley, Cambridgeshire and Lancashire, Lewisham and Northamptonshire, and uses 1995‐1996 as the base year for the information. Inputs were opening hours per week, shelf stock at 31 March 1996, additions to stock in the year, relationship to shopping facilities; demographic variables were total resident population, number of residents unemployed, number over pensionable age and those aged 14 and under; and outputs were issues in the year, visits in the year, and enquiries in the year. This information is used to develop and test a model of how and whether service points were performing better than might be expected from the levels of inputs. If they were, they would serve as examples of good practice within an authority, and so act as benchmarks.

This report is a follow‐up study to an earlier LISU report by Claire Creaser and John Sumsion, Deprivation and Library Performance (LISU Occasional Paper 10, 1995), which investigated the relationship between social deprivation in a local authority area and library use. It related the Department of the Environment′s Index of Local Conditions (HMSO, 1994), to book issues per capita and other library performance indicators and showed that in London and the English metropolitan districts there was a statistical relationship between deprivation and library performance. In this new report on modelling use, Creaser set out to identify the elements that have a significant impact as predictors of lending library use as measured by loans, enquiries and visits per year, and to create a model which could be used to predict probable use of a particular library service point. To achieve this, data was organised and analysed using correlation (for relationships between variables), analysis of variance (for significance of relationship to shopping facilities), and regression analysis (for the predictive model, one variable predicted against another). These are all well‐known techniques in themselves, but, as Creaser rightly says, they have been used all too infrequently in LIS work for purposes of data analysis and forecasting.

Correlations can be good and bad. Some present no surprises and are useful: book issues per capita/book stock per capita (correlation coefficient 0.69 percent, assumed to be Pearson), visits per capita/AV additions per capita (0.40 percent), visits per capita/weekly opening hours (0.45 percent), book stock turn/weekly opening hours (0.37 percent), book issues per capita/percentage unemployment (‐‐0.32 percent), and such data will serve to confirm intuitive professional knowledge about the interaction between such variables. Issues are lower the more unemployed and ethnic minority users there are and the lower the stock turn happens to be, and visits are fewer the lower the total replenishment rate. Scatter diagrams present the data in consolidated form, while tables break down correlations between variables for the six respondent authorities. Other correlations are interestingly low (for example, visits per capita/percentage pensioners at 0.00, AV issues per capita/AV replenishment percent at 0.03, and book stock turn/AV additions per capita at 0.02 percent) or predictably low (and probably remote in any case, such as enquiries per capita/book additions per capita at 0.02 percent). Findings are certainly of interest, even though no close idea of what the original datasets were like and no indication of significance levels for analysis have been provided.

The main thrust of the report is to create and test a model of use at individual service points. The model is based on regression analysis which is a statistical procedure for examining the extent to which an output measure is dependent on one or more input measures (or variables). This is applied five times. The model for book issues, in which predicted book issues per capita are examined against other variables (like total stock, additions, opening hours, percentages of pensioners and children), emerges as a good fit for available data (predicted values are generally close to actual values observed). The model allows us to conclude that if actual book issues are less than half the expected value, the service point could be considered to be performing badly relative to its levels of service provision and the characteristics of its catchment population. Performance categories from good to poor are offered as points of reference for such assessments.

When the model is applied to other variables, like AV issues and visits, enquiries and stock turn, it turns out to be of lower value, however intuitively useful it might appear, so much more work needs to be done to develop and test such models. The relationship individual service points have with shopping facilities, again, has an important intuitive impact on the analysis, and, even though ANOVA was used to compare average levels of issues per capita in the different authorities, and also to highlight differences between authorities, its representation in, say, the visits model (as 1′′ if the library has a poor relationship to shopping facilities and 0′′ otherwise) begs a number of logical and mathematical assumptions. All that said, this report makes fascinating reading, provides interesting correlations, takes care about the validation of its methodology, and provides at least two applications of the regression model that deserve confident development. Statistical analysis, despite the general belief, does not set out to provide answers: it deals in probabilities, it offers indicators, it highlights exceptions, it confirms hunches. Seen and used in this light, utilising methods like the ones Creaser demonstrates here, we are moving closer to an authoritative understanding of what service point performance really is. A work for researchers, practitioners and policy‐makers involved with public library management and ILS decision‐making everywhere.