The relation between the quality of research, researchers’ experience, and their academic environment

Abstract

This article investigates to what extent researchers’ experience and the solidity of their academic environment influence the quality of their research. The hypotheses are derived from the assumptions that experience matters for quality research and that there are great intellectual synergies to be obtained from interacting with many colleagues who are active researchers. All articles published between 2000 and 2006 in five leading transportation journals are included in the analysis, and their research quality is measured by the number of times each article is cited by August 2016. Controlling for other factors influencing citations, such as article age and the number of references, the most important finding is that both experience and academic environment matter for performing quality research. When the authors’ experience, measured by the number of previous publications, increases by 1% from its average level, their published articles are expected to garner 0.31% more citations. Moreover, when the research activity at the unit to which the authors are affiliated, measured by the unit’s total number of publications, increases by 1% from its average level, the number of times their articles are cited will increase by 0.19%. This signals that, relatively speaking, the researchers’ own experience and merits mean more than the academic environment with regard to producing high-quality research. The above results enable us to discuss how researchers’ experience can compensate for working in less active academic communities holding research quality constant.

Appendix: Deducing some mathematical properties of the model used

It follows from Eq. (2) that the expressions for the first and second derivatives are $\frac{{\partial {\text{CI}}}}{\partial i} = \beta_{i} \cdot {\text{CI}} \cdot i^{{\tau_{i} - 1}}$, $\frac{{\partial {\text{CI}}}}{\partial j} = \beta_{j} {\text{CI}}$, $\frac{{\partial^{2} {\text{CI}}}}{{\partial i^{2} }} = \beta_{i} \cdot {\text{CI}} \cdot i^{{\tau_{i} - 2}} \left[ {\beta_{i} \cdot i^{{\tau_{i} }} + \tau_{i} - 1} \right]$ and $\frac{{\partial^{2} {\text{CI}}}}{{\partial j^{2} }} = \beta_{j}^{2} {\text{CI}}$, $i = \left\{ {{\text{PI}},{\text{PP}},{\text{RF}}} \right\},\;j = \left\{ {{\text{AN}},{\text{YP}},{\text{WT}}} \right\}$. When $0 < \tau_{i} < 1$ it implies that $\frac{{\partial {\text{CI}}}}{\partial i},\frac{{\partial {\text{CI}}}}{\partial j} \ge \left( < \right) 0$ when $\beta_{i} ,\beta_{j} \ge \left( < \right)0$, $\frac{{\partial^{2} {\text{CI}}}}{{\partial i^{2} }} > 0$ when $\beta_{i} < {0\,{\text{or}}\,\beta_{i } } > 0\;{\text{and}}\,i > \left[ {\frac{{1 - \tau_{i} }}{{\beta_{i} }}} \right]^{{1/\tau_{i} }} , \frac{{\partial^{2} {\text{CI}}}}{{\partial i^{2} }} < 0$ otherwise. When $\tau_{\text{PI}}$ and $\tau_{\text{PP}}$ are set to the stated values in Eq. (2). $\frac{{\partial^{2} {\text{CI}}}}{{\partial {\text{PI}}^{2} }},\frac{{\partial^{2} {\text{CI}}}}{{\partial {\text{PP}}^{2} }} < 0$ when PI < 278 and PP < 690, whilst $\frac{{\partial^{2} {\text{CI}}}}{{\partial {\text{RF}}^{2} }} < 0$ when RF < 2.54. Moreover, $\frac{{\partial^{2} {\text{CI}}}}{{\partial j^{2} }} > 0$ for all values of $\beta_{j}$. The derivatives show that neither are constant but rather vary with the values of $i$ and j. The absolute change in ${\text{CI}}$ when one variable changes by one unit is thus dependent on the values of all explanatory variables in the model.

To obtain a good interpretation of the $\beta$-values, we can focus on the elasticity values. The above derivatives imply that ${\text{EL}}_{i} {\text{CI}} = \frac{{\partial {\text{CI}}}}{\partial i}\frac{i}{\text{CI}} = \beta_{i} \cdot i^{{\tau_{i} }}$ and ${\text{EL}}_{j} {\text{CI}} = \frac{{\partial {\text{CI}}}}{\partial j}\frac{j}{\text{CI}} = \beta_{j} \cdot j$ where ${\text{EL}}_{i} {\text{CI}}$ and ${\text{EL}}_{j} {\text{CI}}$ denote the elasticity of ${\text{CI}}$ with respect to $i$ and $j$, respectively. Hence, the absolute values of the elasticities of ${\text{CI}}$ with respect to $i, i = \left\{ {{\text{PI}},{\text{PP}},{\text{RF}}} \right\}$ increase concavely with i, whereas the elasticities of ${\text{CI}}$ with respect to $j, j = \left\{ {{\text{AN}},{\text{YP}}, \ldots } \right\}$ increase proportionally with $j$.

From the above derivatives it also follows that $\left( {\frac{{\frac{{\partial {\text{CI}}}}{\partial i}}}{\text{CI}}} \right) = \beta \cdot i^{{\tau_{i} - 1}}$ and $\left( {\frac{{\frac{{\partial {\text{CI}}}}{\partial j}}}{\text{CI}}} \right) = \beta_{j} .$ Thus, an increase in i by one unit will always increase ${\text{CI}}$ by $\left( {100 \cdot i^{\tau_{i} - 1} \cdot \beta_{i} } \right) \%$, while an increase in $j$ by one unit will always increase ${\text{CI}}$ by approximately ($100 \cdot \beta_{j} )$%. Our model specification implies that the relative changes in ${\text{CI}}$ when $i$ and j change by one unit, are independent of the values of the other explanatory variables. Moreover, the relative changes in j, $j = \left\{ {{\text{AN}},{\text{YP}},{\text{WT}}} \right\}$ are independent of their own values whereas the relative changes in i, $i = \left\{ {{\text{PI}},{\text{PP}},{\text{RF}}} \right\}$ decrease with their own values.

Equation (2) implicitly defines ${\text{PP}}$ as a function of ${\text{PI}}$ for given values of ${\text{CI}} ({\text{CI}}^{*} )$ and of the other explanatory variables (${\text{AN}}^{*} ,{\text{RF}}^{*} ,{\text{WT}}^{*} ,{\text{YP}}^{*} ,{\text{TR}}^{*} ,{\text{TA}}^{*} ,{\text{TB}}^{*} ,{\text{TS}}^{*} )$; that is

$${\text{PP}} = F\left( {\text{PI}} \right) =\left[\frac{{\ln {\text{CI}} - \beta_{PI} {\text{PI}}^{{\tau_{\text{PI}} }} - q}}{{\beta_{\text{PP}} }}\right]^{{1/\tau_{\text{PP}} }}$$

in which q = $\beta_{0} + \beta_{\text{RF}} {\text{RF}}^{0.37*} + \beta_{\text{AN}} {\text{AN}}^{*} + \beta_{\text{WT}} {\text{WT}}^{*} + \beta_{\text{YP}} {\text{YP}}^{*} + \beta_{\text{TR}} {\text{TR}}^{*} + \beta_{\text{PA}} {\text{PA}}^{*} +\, \beta_{\text{PB}} {\text{PB}}^{*} + \beta_{\text{TS}} {\text{TS}}^{*}$

From the F(PI) relationship above we can deduce the marginal rate of substitution $\left( {\text{MRT}} \right)$ between ${\text{PP}}$ and ${\text{PI}}$:

$${\text{MRT}} = \frac{{\partial {\text{PP}}}}{{\partial {\text{PI}}}} = - \frac{{\beta_{\text{PI}} }}{{\beta_{\text{PP}} }} \cdot \frac{{\tau_{\text{PI}} }}{{\tau_{\text{PP}} }} \cdot \frac{{{\text{PP}}^{{1 - \tau_{\text{PP}} }} }}{{{\text{PI}}^{{1 - \tau_{\text{PI}} }} }}$$