The prediction model was developed using binary logistic regression, which is used to identify the relationship between one or more predictor variables and a categorical outcome variable [17] in a univariate or multivariate approach. In this paper the binary logistic regression is used to analyze the influence of 18 factors (predictor variables, see ) over the dichotomous outcome variable *system preference* (LN or ELN). Unlike other statistical methods the “(…) logistic regression analysis does not require that the data are drawn from a multivariate normal distribution with equal variances and covariances for the explanatory variables.” [18, page 1390].

Table 1 Univariate logistic regressions for each individual variable for pre-selection.

In literature a minimum sample size between 50 and 100 is recommended [5, 23]. The required size increases with an increasing number of predictor variables. Considering this, a pre-selection of relevant variables was performed. In this first step each predictor will be individually tested for statistical significance in a univariate logistic regression. In order to include all influencing factors the predictors with a significance level of p≤20% in the unvariate model are tested for significance in a multivariate model. In the second step, insignificant predictors with p>5% are excluded iteratively from the prediction model (see Figure 2).

Figure 2 Structure chart of the methodology.

Model fit in a logistic regression is sensitive to collinerarity among the predictors [8]. Therefore, the dependence of the predictor variables is examined and considered in the model. Finally, validation tests for the overall model and goodness-of-fit are executed (overall statistics, Hosmer and Lemeshow test, Nagelwerke R^{2}, Omnibustest). All statistical calculations were carried out using SPSS (v.19).

The survey participants were asked about their professional backgrounds (multiple replies possible). The free text answers were categorized into the fields of engineering, natural sciences and human medicine based on the *personnel at universities* subject classification [22].

The variable *affinity for technology* is based on five questions about the frequency of using laptops, tablet PCs, smartphones, social networks and Internet forums. In order to identify more and less technically able researchers, the answers were weighted with never=0, less often than weekly=1, weekly=2 and daily=3. The cumulative results of these five questions reflect the level of affinity for technology.

Another issue, the anticipated strengths of the systems LN and ELN, was analyzed in ten different categories (see ). For each category the question included three answer categories:

LN has an advantage over ELN

Neither LN nor ELN has an advantage

ELN has an advantage over LN

It is to be expected that factors determining the researcher’s system choice (LN/ELN) are also seen as strength of the system. In order to interpret the factors with regard to the low rate of ELN in the academic environment, the answers (2) and (3) are combined to “LN has no advantage over ELN”. In the logistic regression analysis, missing answers are excluded from analysis.

## Comments (0)

General note:By using the comment function on degruyter.com you agree to our Privacy Statement. A respectful treatment of one another is important to us. Therefore we would like to draw your attention to our House Rules.