Modelling a Causal Relationship between the Internet and Academic Research Performance in an Australian University: A Case Study

Abstract

This study examines the relationship between academics’ use of the Internet for academic purposes and their research performance using cross-sectional data collected from academics of the University of Southern Queensland (USQ), Australia, during the period February–March 2014. In this study, a system of simultaneous equation models is used to control the potential bias associated with simultaneity between the use of the Internet and academics’ research performances. The simultaneity, a potential econometric problem, was overlooked in past studies. A finding of this study is that academics’ use of the Internet is a statistically significant contributor to research output in an Australian university. The estimated elasticity of research output is 0.16 with respect to changes in Internet use.

JEL Classifications: C3, D8, I2

Keywords

Australia Internet Research Production Simultaneous Equation Model

1. INTRODUCTION

The question of what the determinants of (academic) research productivity are has been a basic concern for long, and has resulted in many research papers over the last 20 years. Researchers have found various determinants that include gender (Bently, 2011; Jung, 2012; Padilla-Gonzalez et al., 2011), ethnicity (Mamiseishvili and Rosser, 2010; Webber, 2011), academic rank (Lissoni et al., 2011; Mishra and Smyth, 2013), research experience (Fukuzawa, 2014; Jung, 2012), institutional features (Dundar and Lewis, 1998; Iqbal and Mahmood, 2011), research collaboration (Abramo et al., 2008; Kartz and Martin, 1997), availability of research grants (Edgar and Geare, 2011) and research management (Jordan et al., 2013). However, the aforesaid studies have not included information and communication technologies (ICTs), particularly the use of the Internet by academics, to model the research production function, even though in education the contribution of ICTs is well documented in some literature. Scholars (e.g., Jovanovic and Rouaaeau, 2005) argue that ICTs are an enabling technology, and as such influence productivity¹ positively and tremendously in manufacturing and servicess (see Cardona et al., 2013, for a survery of literature). In education, the contribution of ICT to teaching productivity has been examined in some literature too (Murphy et al., 2013). Therefore, establishing a research evidence empirically on the relationship between the academic use of the Internet and academic research output is meaningful. This article examines the relationship in a regional Australian university—the University of Southern Queensland (USQ), Australia—using cross-section data on academics working full-time during the period January–February 2014.

It is argued in the literature that researchers in developed countries are increasingly reliant on Internet facilities for research and national development (Okafor et al., 2011). The growth of the Internet, particularly broadband Internet powered by high-speed bandwidth, has increased the potential for growth of academic research in developed countries, including Australia. The Australian government has been expanding its National Broadband Network (NBN) infrastructure facilities connecting every house and institution and as a result, access to the Internet has been increasing.

The main contribution of this article is methodological, as it applies a new model: a system of simultaneous equations and estimation techniques. The other contributions are that it is the first-ever study conducted in Australia, where broadband Internet has been expanding rapidly for the last couples of years. The findings of this research may be used as a future reference by researchers.

The rest of the article is organised into the following sections. The second section presents the background and literature review. The third and fourth sections present the data and econometric model. The fifth section presents the results and discussion. This article ends with conclusions and limitations.

2. BACKGROUND

2.1 An Overview

The literature argues that many academics use the Internet for academic purposes, such as general information gathering, background, context study, analyses of data collected from the website and website study (Allen et al., 2006). The academic use of the Internet contributes to scholarly activities through diverse mechanisms such as participation in electronic discussion, access to online databases, libraries, electronic journals and newsletters (Organ and McGurk, 1996). Nowadays, almost all academic institutions store scholarly resources on websites, which are accessible either commercially or free of cost. Universities subscribe to the scholarly research databases such as Science Direct and JSTOR. Consequently, the demand for commercial research databases, such as bibliographic databases (formerly only available on CD-ROM), is increasing. Further, the number of electronic journals is increasing continuously, as are online publication opportunities. Over the past couple of years, journal publication (a measure of research output) has increased substantially (Figure 1). As a whole, the Internet has changed the nature of scholarly outputs.

Figure 1

Source: Authors’ presentation based on Australian Department of Education data. Available at http://docs.education.gov.au/node/34561

2.2 Previous Literature

Our survey shows that the literature on research productivity is voluminous. Previous studies have identified the determinants of research output broadly divided into three categories: personal attributes (researcher’s sex, age, education, etc.), institutional and departmental attributes (characteristics of the institution, size of faculty, technology and instrumental infrastructures available, etc.) and environmental attributes (labour policies, public and private funds available, students available to support the research, etc.). A survey of literature (presented in Appendix 1) has found various determinants of research productivity, as well as a very diverse pattern in the relationship between the inputs and outputs. Further, in most of the studies, the researchers have modelled the relationships assuming that all the right-hand side variables are strictly exogenous. The endogenous effects of some of the variables have not been considered, even though a study has examined the possibility of the influence of research productivity on research collaboration (Lee and Bozeman, 2005), which indicates a likely simultaneous relationship between inputs and outputs in research production studies. Therefore, in our study, we relax the exogeneity assumption of the right-hand side variables—research collaboration and the Internet—within the framework of a simultaneous equation model and estimate the effect of Internet use on research output.

3. METHODOLOGY

3.1 Area of Study

The study area is the University of Southern Queensland (USQ), a regional government university. According to university data, between 2010 and 2013, the average number of full-time academic staff increased by 3.87 per cent. On the other hand, according to the electronic depository account between 2001 and 2013, research output, measured by publications in peer-reviewed journals, conference proceedings, book chapters and books, increased by 87 per cent (i.e., total research outputs which were around 10 in 2001 increased to over 70 in 2012) (for details, refer to http://eprints.usq.edu.au). But in 2013 this decreased to 50, which might be attributed to the cut in federal government funding to Australian universities, including research funding to government universities with a view to financing the Gonski reform, a reform measure undertaken by the past Australian government to finance primary and secondary education. Between 2008 and 2009, the Australian government spent 0.53 per cent of its per capita gross domestic product (GDP) on R&D, a share that continued to rise (Australian Bureau of Statistics, 2008), but in a global context, this is lower than the amount spent in the US, Japan, Germany, Finland and the Republic of Korea on R&D (Figure 2).

Figure 2

Source: Authors’ calculation based on UNSECO data: Science, Technology and Innovation. Available at http://data.uis.unesco.org/Index.aspx?DataSetCode=SCN_DS (accessed on 20 June 2014).

3.2 Data Type and Collection

The total population of academic staff at the time of data collection was 466, and not all of them were teaching academics. Academic non-teaching staff remained outside the population of our survey. Further, all of them were not full-time academics; 87 per cent of them (i.e., 391 academics) were full-time and the remaining were casual and contractual academics who were not part of the population of our survey data (see Appendix 2 for questions asked for data collection).

To collect the data, an online survey instrument was developed and administered to the participants. The data collection instrument was prepared based on past literature and in consultation with research experts in the USQ. The instrument consisted of both open-ended and closed-ended questions: the continuous variables were open-ended questions and the categorical variables were closed-ended questions.

Before finalising the survey instrument, the data collection instrument was piloted three times among randomly selected potential participants to ensure the highest quality of the instrument. The online data collection instrument was circulated among the participants using an online survey platform: Qulatrics. Approximately 6 weeks were given for the academics to participate in the survey. A total of 83 responses were received, giving a participation rate of approximately 21 per cent. Of these, only 63 responses are usable. Among the respondents, 55 per cent were males and 45 per cent were females. In terms of academic rank, 6 per cent were associate lecturers, 45 per cent were lecturers, 28 per cent senior lecturers, 11 per cent were associate professors and 11 per cent were professors.

4. ECONOMETRIC MODELS

The education production function (Cohn and Geske, 1992; Hanushek, 1986) examines the relationship between inputs and output in the educational process. The process relies on quantitative investigations based on econometric methods (Hanushek, 1986). In higher education institutions, particularly universities, the education output is different from the education output in schools because universities are referred to as multi-product firms and produce multiple outputs (Cohn et al., 1989; Mamun, 2012). Schools, on the other hand, are referred to as single-product firms and produce a homogenous product (Belfield, 2000). Research is one of the multiple outputs in a university, the other outputs are undergraduate, graduate students and public services (see Cohn et al., 1989 for details). Theoretically, research output is linked to various inputs including researchers’ personal and institutional characteristics (Lewis and Dundar, 1995), and firm size (Abbott and Doucouliagos, 2004).

One strong assumption about an education production function is that all the right-hand side variables are exogeneous and the left-hand variable is endogeneous. In our study, we relax the strong assumption and assume that some of the right-hand side variables are endogenous. For instance, the use of the Internet by academics is endogenous because of the influences of unobserved and observed exogenous variables such as users’ age and level of education. Research collaboration, which is a righ-hand side variable in a research production function, is endogeneous because of our assumption that research output and resarch collaboration have a bidirectional relationship (study by Lee and Bozeman, 2005). This phenomena is known as the simultaneity problem in econometrics. Our argument is that participation in a research collaboration is a matter of personal choice. Collaboration requires a connection among researchers, which generates social capital, and investment in the social capital is a matter of an individual researcher’s choice, which is influenced by some unobserved factors too. Wissen and Golob (1990) have asserted that an individual makes a decision not in isolation but in conjunction with a number of circumstances and conditions. It is likely that a researcher’s academic position, institutional support, research strengths and weaknesses, availability, communication skills and reputation in the research area may also influence collaboration. Therefore, the association between research collaboration and research output is not free from selection bias. Because of the influence of the selection bias, collaborative research may be an endogenous variable in single-equation model. In such a circumstance, the simultaneous equation model is suitable to estimate econometrically the unbiased estimates of the parameters (Cameron and Trivedi, 2005; Greene, 2012). Consequently, a test of the exogeneity hypothesis will be required to understand the endogeneity hypothesis of the two variables: Internet use and research collaboration.

Table 1
Endogeneity Test Results

Variables Durbin (score) Wu–Hausman

Internet $x_{(1)}^{2}$ = 4.21; p = 0.04 F(1,63) = 4.22; p = 0.04

Collaborative research $x_{(1)}^{2}$ = 13.86; p = 0.00 F(1,63) = 16.44; p = 0.00

Variables	Durbin (score)	Wu–Hausman
Internet	$x_{(1)}^{2}$ = 4.21; p = 0.04	F(1,63) = 4.22; p = 0.04
Collaborative research	$x_{(1)}^{2}$ = 13.86; p = 0.00	F(1,63) = 16.44; p = 0.00

Source: Authors’ calculations.

We conduct the Hausman–Wu test of the exogeneity hypothesis for the two endogenous variables separately: the Internet use variable and participation in collaborative research. Here, the null hypothesis is that both Internet use and participation in collaborative research are exogenous variables. The test statistics are significant at 5 per cent level, implying that both variables are endogenous in the research production function (Table 1).

Theoretically, the endogeneity occurs because academics who are engaged in collaborative research work spend more time on the Internet to communicate with their colleagues for research-related work and vice versa.

In our study, the simultaneous equation model consists of three functions: the research production function, the Internet-use function and the research collaboration function. Each function is linked to the theory and past research evidences which are discussed in the following sections. Based on past study evidences and theory, this study is specified as

y_{i} = α_{0} A_{i} + α_{1} G_{i} + α_{2} I_{i} + α_{3} T_{i} + α_{4} Q_{i} + α_{5} C_{i} + ε_{1}

(1)

Equation (1) describes the research production function, where the right-hand side variable y_i is diverse and includes journal articles, book chapters, books, conference papers, patents, artworks, public lectures and more (Carrington et al., 2005). The right-hand side variables are A_i denoting the age of an academic, G_i denoting research grants, I_i denoting average weekly Internet use by an academic, T_i denoting teaching load of an academic, Q_i denoting academic qualification of an academic, C_i denoting participation in collaborative research work and ε₁ denoting an error term.

Because of the diversity of the dependent variable, a common measure of research output combining different measures of research outputs is problematic. Research quantum and a weighted composite index are two potential measures used in previous studies to measure research outputs at the institutional and individual levels. Carrington et al. (2005) have used a ‘weighted publications’ index in the past that includes books, book chapters, journal articles and conference papers. Following this, we use a weighted composite index too, which includes books, book chapters, journal articles and conference papers. We use the Commonwealth Scholarship Guidelines to Research² to assign weights: we have given a weight equivalent to 1 for journal articles, 1 for conference papers and 1 for book chapters and 5 for a full-length research book and thereafter calculate a weighted average of total research publications. If the total calculated value is less than 1 but greater than 0, we give a default value equal to 1. Moreover, any fractional value is converted to the nearest round number. For example, 1.2 and 1.5 are converted to 1 and 2, respectively. In our dataset, approximately 5 per cent of the data required conversion to the nearest round number.

A frequency distribution of the dependent variables shows that there is a tail in the distribution (Figure 3), which indicates that the majority of academics have a weighted average of less than 2.5. In order to normalise the distribution, we take the natural logarithm of the data. As per the data, 34.32 per cent of the academics have a average (weighted) research output of around 1 and 55 per cent have an average (weighted) research output within the range of 1.1–5.0. A small proportion has an average (weighted) output of over 5.0.

Figure 3

Source: Data set used for this study.

The right-hand side control variables are selected based on research evidences found in previous studies. Meyer (2012), Iqbal and Mahmood (2011) and Jung (2012) have found evidence of the influence of the academic teaching load in the US, Pakistan and Hong Kong in their studies. Based on this, teaching workload is selected as a potential determinant in Equation (1). In our survey, we have found that only 3 per cent of the academics have a full-time teaching load (i.e., 91–100 per cent of the total workload).

The importance of a research grant is evidenced in previous studies by Abbott and Doucouliagos (2004), Teodorescu (2000), Lee and Bozmar (2005) and Bently (2011). Following this, we have incorporated research grants in Equation (1). In our survey, we have found that 45 per cent of academics have not secured any research grants. Among grant recipients, male academics have received an average of Australian dollars (A$) 239,000 and female academics have received an average of A$89,000. The independent t-test shows that the mean difference between male and female grant recipients is not statistically significant at 5 per cent level (t-value –1.15; df = 35).

Other control variables—age, academic rank and academic qualifications—are also included in Equation (1) based on past research evidence. Though the difference between males and females is found to be statistically insignificant in Bently’s study (2001), the variable is found to be statistically significant in studies by Lissoni et al. (2011) and Over (1982). In the context of the given debate, we have included age as a determinant of research output in Equation (1). Some studies have highlighted academic seniority and doctoral qualification of academics as important determinants of research productivity (Bently, 2011; Fukuzawa, 2014; Ho, 1998; Mishra and Smyth, 2013; Ramsden, 1994). Further, the role of ethnicity of academics is highlighted in very recent studies (Edgar and Geare, 2011; Webber, 2011). Webber (2011) has classified ethnicity into US-born academics and foreign-born academics, and Edgar and Geare (2011) have classified ethnicity into European New Zealand, European, New Zealand Maori, Asian and others in their respective studies. In our study, the ethnicity variable is classified according to the first language of the academic: English-speaking academics and others. Here, our assumption is that if the first language of an academic is English, it implies that the academic is an Australian-born or European English-speaking country-born academic. Our dataset shows that Asian-born academics come from Bangladesh, India, Sri Lanka and China.

Equation (1) is an extended model of previous studies where we have incorporated time spent by academics on the Internet for their academic work. In our survey, we have found that 43 per cent of academics use the Internet for 46 hours per week or more and 15 per cent use it for 16–30 hours. Further, female academics use the Internet for 46.75 hours per week on average and male academics for 39.85 hours per week on average. An independent t-test was carried out to test whether there is any significant difference between male and female academics regarding their use of the Internet. The estimated t-test value is 1.26 (df = 65). This test result does not reject the null hypothesis that the difference between male and female academics is equal to 0 at 5 per cent level.

In order to explore the actual functional relationship between the main explanatory variable of interest (i.e., use of the Internet) and the dependent variable (i.e., research output) without imposing any assumptions on their functional relationship, we run a locally weighted scatterplot smoothing (LOWESS), a non-parametric regression (local mean smoothing) (Figure 4). The LOWESS curve shows a linear increasing relationship between the two series of data: the log of weekly Internet use measured in hours (x-axis) and the log of research output (y-axis) by weighted regression line.

Figure 4

Source: Authors’ own.

Now, we move to specify the determinants of average weekly Internet use by academics. The consumption model (CM) of microeconomics and the technology acceptance model (TAM) of technology diffusion are two theories used by researchers to interpret the diffusion of technology in the past. For example, in a study, Seyal et al. (2002) have found that perceived usefulness and perceived easiness are two determinants of Internet use by academics in Brunei Darussalam. However, the perceived benefit of the use of the Internet is unobservable and difficult to measure. Zhang (2013) combines both the CM and the TAM and introduces the Internet consumption model that states that, along with perceived usefulness and perceived easiness, Internet affordability is a key determinant of Internet use at the individual level. As a proxy for academics’ perceived benefit from the use of the Internet, we use research output (or publication). As our study is carried out in an institutional setting where academics’ access to the Internet is free of cost, affordability of the Internet is not a concern here. Moreover, following the International Telecommunication Union (ITU) Report 2011, we use individual education level, income level, gender, age and location as potential determinants of Internet use. So, the Internet use model is

I_{i} = β_{0} A_{i} + β_{1} y_{i} + β_{2} W_{i} + β_{3} Q_{i} + ε_{2}

(2)

Equation (2) describes the determinants of the use of the Internet by academics. It incorporates two variables which are different from those in Equation (1): W_i = dummy variable for sex of an academic (0 = male; 1 = female) and ε₂ which is the error term for Equation (2).

The contribution of research collaboration to research production is well established in literature (Abramo et al., 2008; Bently, 2011; Lee and Bozeman, 2005; Lissoni et al., 2011). In spite of using different methodologies, the studies have found a positive influence of international or extramural research collaboration on research output. Based on previous study evidences, we have developed Equation (3) that describes the determinants of collaborative research work in our study. Here, the determinants are R_i denoting rank or position of an academic, N_i denoting native language of an academic and ε₃ denoting an error term.

C_{i} = γ_{0} y_{i} + γ_{1} R_{i} + γ_{2} N_{i} + ε_{3}

(3)

In our study, equations (1)–(3) constitute a simultaneous equation system in which y_i, I_i and C_i are endogenous variables, and α₀ – α₅, β₀ – β₃ and γ₀ – γ₃ are coefficients to be estimated. We are interested in the coefficient of α₂. The variable coefficients explain intermediate or proximate influence of the Internet (I) on research output (y). The reduced-form coefficients provide us the equilibrium impact.

The summary statistics (Table 2) indicate that around 43 per cent of the academics on average use the Internet for over 46 hours a week, around 80 per cent have a PhD degree and around 45 per cent hold the rank of senior lecturer. The identification condition for simultaneous equations is satisfied because a different set of independent variables is included in equations (1), (2) and (3) (Table 3). This implies that each equation contains some variables that do not appear in the other equations.

Table 2

Summary Statistics of the Categorical and Continuous Variables

	Frequency	Percentage
Ethnicity
Others	18	26.87
English	49	73.13
Academic Qualifications
Bachelors degree	3	4.48
Masters degree	10	14.93
PhD	54	80.60
Rank
Associate lecturer	4	5.97
Lecturer	30	44.78
Senior lecturer	19	28.36
Associate professor	7	10.45
Professor	7	10.45
Internet Use
1–15 hours	9	13.43
16–30 hours	17	25.37
31–45 hours	12	17.91
46 hours and above	29	43.28
Age
Less than 34	12	17.91
35–49	30	44.78
50–64	22	32.84
Above 65	3	4.48

Continuous Variables
	Unit	Mean	Std Dev.	Min.	Max.
Research grant	1,000	102.2985	300.5395	0	1740
Internet use	Weekly hours	42.73	22.14	7	112

Source: Authors’ calculations.

Table 3

Model Specification

Endogenous Variables
Research output	Log of average research output
Collaborative research	Self-assessed participation in collaborative research. 1 = yes; 0 = no
ln (Internet)	Log of weekly Internet use (hours)
Common variables in equations (1), (2) and (3)
Age	Age (continuous variable)
ln (grant)	Log of research grant (A$)
Academic qualification	1 = PhD degree holder; 0 = otherwise
Additional variable appeared in Equation (1)
ln (teaching)	Log of teaching load (per cent)
Additional variable appeared in Equation (2)
Sex	Dummy; 1 = male; 0 = female
Additional variable appeared in Equation (3)
Rank	Dummy; 0 = associate lecturer; 1 = lecturer; 2 = senior lecturer; 3 = associate professor; 4 = professor
Native language	Dummy; 1 = English; 0 = otherwise

Source: Authors’ calculations.

In order to impose the normality condition, we take the natural log of the selected variables that has a skewed distribution. However, in order to avoid dropping any observation that has a value of 0, such as research output, following the suggestion put forward by Wooldrige (2000), we transform 0 to 1, before taking the natural logarithm. So, the system of simultaneous equations consisting of log-linear equations is

\begin{matrix} \begin{matrix} \ln ​ (y) ​_{i} = α_{0} A ​_{i} + α ​_{1} \ln ​ (G) ​_{i} + α ​_{2} \ln ​ (I) ​_{i} + α ​_{3} \ln ​ (T) ​_{i} + α ​_{4} ​ Q_{i} + α ​_{5} ​ C_{i} + ε ​_{1} \end{matrix} \\ \begin{matrix} \ln ​ {(I)}_{i} = β ​_{0} A_{i} + β ​_{1} \ln ​ (y) ​_{i} + β_{2} ​ W_{i} + β ​_{3} ​ Q_{i} + ε ​_{2} \\ C_{i} = γ ​_{0} \ln ​ (y) ​_{i} + γ ​_{1} ​ R_{i} + γ ​_{2} N_{i} + ε ​_{3} \end{matrix} \end{matrix}

(4)

In order to estimate a system for Equation (4), two types of estimators are considered: the two-stage estimator (2SLS) and the three-stage estimator (3SLS) (Gujarati, 2003; Wooldgride, 2002). The literature suggests that the former is cheaper computationally, while the latter is efficient asymptotically. It is suggested, further, that if the 3SLS and 2SLS estimates are significantly different from each other, then one should consider the 2SLS. A similar suggestion has also been put forward by other literature (Gujarati, 2003; Wooldridge, 2002) because in the presence of simultaneity, 2SLS will give consistent and efficient results if all the equations are correctly specified. Under such circumstances, the Hausman test is suggested to test the difference in coefficients between the two estimators (Spencer and Berk, 1981).

5. RESULTS AND DISCUSSIONS

First, we begin with reporting the results of the over-identification test, independence test and heteroskedasticity test. The Hansen–Sargan over-identification statistic is $χ_{(4)}^{2} = 2.48$ , p-value = 0.65. The test results confirm that the exclusion restrictions are valid in our study. The Breusch–Pagan test of independence is $χ_{(1)}^{2} = 0.165$ , p-value = 0.68. The statistic has a very large p-value, implying that the test result does not reject the null hypothesis of dependency (the alternative hypothesis is independency). This implies that the error terms of the three equations in the system of equations are correlated; thus, it is required for a joint determination of the system of equations. Finally, the Breusch–Pagan/Cook–Weisberg test for heteroskedasticity is very small (i.e., 0.32; df = 1) and the estimated p-value is 0.57. Therefore, the null hypothesis is not rejected at the 5 per cent level. This implies that there is no heteroskedasticity in our data set.

At the very outset, we have estimated the system of equations by the 2SLS estimator (column 3, Table 4 labelled Model (1)). Next, we have estimated the same model by 3SLS (column 2, Table 4). In order to compare the efficiency of the two estimates, we conducted a Hausman test and reported the results at the bottom of the table. The estimated Hausman test statistic is 5.52 with a p-value of 0.99; because of the high p-value the null hypothesis is not rejected, which means that the difference in the coefficients between the two estimators is not systematic. Hence, the 2SLS estimator is used as an efficient estimator to derive unbiased estimates of the parameters of interest in the remaining parts of this article.

Table 4
Estimation of the Simultaneous Equation Model

Sensitivity Analysis

2SLS (1) 3SLS (2) Bootstrap (3) Without a Variable (4)

ln (res. output) Age –0.002(0.38) –0.010(2.88) –0.001(0.35) –

Panel A ln (grant) 0.033 0.013 0.033 0.033

(3.39) (1.94) (3.79) (3.44)

ln (Internet) 0.164 0.330 0.16 0.149

(2.12)* (6.93)** (2.26)* (2.25)*

ln (teaching) 0.003 –0.002 0.003 0.001

(0.09) (0.10) (0.11) (0.02)

Qualification (0 = otherwise)

1 = doctorate 0.295 0.138 0.30 0.289

(2.27)* (1.20) (2.57)* (2.27)*

Collaboration (0 = no)

1 = yes 0.096 0.170 0.096 0.082

(0.50) (1.60) (0.61) (0.44)

R-square

0.88
0.85
0.85
0.89

ln (Internet) ln (res. output) 2.147 2.079 2.15 2.147

Panel B (4.00) (4.31) (4.15) (4.00)

Age 0.034 0.033 0.034 0.034

(3.60) (3.75) (3.61) (3.60)

Sex

Male 0.008 –0.021 0.008 0.008

Qualification (0 = otherwise) (0.03) (0.13) (0.03) (0.03)

1 = doctorate –0.341 –0.163 –0.34 –0.341

(0.78) (0.40) (1.00) (0.78)

R-square

0.91
0.90
0.90
0.90

Collaboration ln (res. output) 0.17 0.170 0.17 0.166

(1.59) (1.75) (1.26) (1.59)

Panel C Rank 0 = associate lecturer

1 = lecturer 0.681 0.679 0.68 0.681

(5.95) (6.47) (3.74) (5.95)

2 = senior lecturer 0.659 0.668 0.66 0.659

(5.01) (5.55) (3.24) (5.01)

3 = associate professor 0.699 0.712 0.70 0.699

(4.78) (5.33) (3.82) (4.78)

4 = professor 0.680 0.711 0.68 0.680

(4.36) (4.99) (3.36) (4.36)

Native language 0 = otherwise

1 = native 0.157 0.141 0.15 0.157

(2.93) (2.90) (1.79) (2.93)**

0.96 0.97 0.96 0.95

R-square

Hausman test3SLS vs 2SLS Chi-square (16) = 5.52; p-value = 0.99

N 67 67 67

		Sensitivity Analysis
ln (res. output)	Age	–0.002(0.38)	–0.010(2.88)**	–0.001(0.35)	–
Panel A	ln (grant)	0.033	0.013	0.033	0.033
		(3.39)**	(1.94)	(3.79)**	(3.44)**
	ln (Internet)	0.164	0.330	0.16	0.149
		(2.12)*	(6.93)**	(2.26)*	(2.25)*
	ln (teaching)	0.003	–0.002	0.003	0.001
		(0.09)	(0.10)	(0.11)	(0.02)
	Qualification (0 = otherwise)
	1 = doctorate	0.295	0.138	0.30	0.289
		(2.27)*	(1.20)	(2.57)*	(2.27)*
	Collaboration (0 = no)
	1 = yes	0.096	0.170	0.096	0.082
		(0.50)	(1.60)	(0.61)	(0.44)
R-square		0.88	0.85	0.85	0.89
ln (Internet)	ln (res. output)	2.147	2.079	2.15	2.147
Panel B		(4.00)**	(4.31)**	(4.15)**	(4.00)**
	Age	0.034	0.033	0.034	0.034
		(3.60)**	(3.75)**	(3.61)**	(3.60)**
	Sex
	Male	0.008	–0.021	0.008	0.008
	Qualification (0 = otherwise)	(0.03)	(0.13)	(0.03)	(0.03)
	1 = doctorate	–0.341	–0.163	–0.34	–0.341
		(0.78)	(0.40)	(1.00)	(0.78)
R-square		0.91	0.90	0.90	0.90
Collaboration	ln (res. output)	0.17	0.170	0.17	0.166
		(1.59)	(1.75)	(1.26)	(1.59)
Panel C	Rank 0 = associate lecturer
	1 = lecturer	0.681	0.679	0.68	0.681
		(5.95)**	(6.47)**	(3.74)**	(5.95)**
	2 = senior lecturer	0.659	0.668	0.66	0.659
		(5.01)**	(5.55)**	(3.24)**	(5.01)**
	3 = associate professor	0.699	0.712	0.70	0.699
		(4.78)**	(5.33)**	(3.82)**	(4.78)**
	4 = professor	0.680	0.711	0.68	0.680
		(4.36)**	(4.99)**	(3.36)**	(4.36)**
	Native language 0 = otherwise
	1 = native	0.157	0.141	0.15	0.157
		(2.93)**	(2.90)**	(1.79)	(2.93)**
		0.96	0.97	0.96	0.95
R-square
Hausman test3SLS vs 2SLS	Chi-square (16) = 5.52; p-value = 0.99
N		67	67	67

Source: Authors’ calculations.

Note: * denotes p < 0.05 and ** denotes p < 0.01.

We analysed the sensitivity of the system of equations to variations in small size and the exclusion of a variable. We applied the bootstrapping technique to check sensitivity with respect to small size. The bootstrap technique is a slightly different type of Monte Carlo simulation, used by researchers to test the hypothesis in empirical research (Cameron and Trivedi, 2005). We simulated the model by applying bootstrapping 200 times in order to test the hypothesis of our study; 200 times is the standard rule of thumb proposed by Efron and Tibsharani (1993, p. 52). The estimated coefficients in (3) are not significantly different from the coefficients in (1) (Table 4). The sign of the coefficients is not changed at all.

Next, in order to understand sensitivity to the exclusion of variable, we drop the age variable from the research production equation and re-estimate the model [labelled (4) in Table 4]. The estimated results do not vary substantially from the results labelled (1) in Table 4.

5.1 Determinants of Research Output

In our analysis, we have found that three of the five explanatory variables are statistically significant at the 5 per cent level (Panel A, Table 4). These variables are the log of research grants (ln grant), the log of weekly Internet use (ln Internet) and the dummy variable for academic qualification (Qualification).

The estimated coefficient on research grants is 0.033 and is statistically significant at the 5 per cent level. The estimated coefficient is found to be consistent in all estimators. The variable has a positive sign, which implies that the research grant is an important determinant that influences research output positively. This is consistent with the findings by Abbott and Doucouliagos (2004) and Bently (2011), who have also found a positive effect of a research grant on research output.

The estimated coefficients of the teaching and collaborative research variables are found to be positive but statistically insignificant. This finding is in contrast with the findings by Jung (2012) and Iqbal and Mahmood (2011). Jung (2012) and Iqbal and Mahmood (2011) have found a negative influence of teaching load on research in Hong Kong and Pakistan. Our finding implies that teaching load is an integrated part of the overall academic responsibilities of an academic. There is a clear distribution of time between the two categories of work, and time allocated for research does not intervene on time allocated for teaching.

Regarding the effect of research collaboration, our study finding is in contrast with study findings by Abramo et al. (2008) and Bently (2011). Bently (2011) has found that international research collaboration has a statistically insignificant effect on research output in Australian universities. The difference in findings might be because of the differences in the econometric model used in our study and in Bently’s study. The estimated coefficient for age is found to be small and the sign is negative. The sign of the coefficient is consistent with the study by Lissoni et al. (2011), but unlike Lissoni et al. (2011), we have found an insignificant effect of age on research output.

The estimated coefficient of academic qualification is 0.29, which is statistically significant at the 5 per cent level. The positive sign of the coefficient implies that academics who have a doctorate make a higher impact on research output compared to academics without a doctoral degree. This is an expected result because a doctoral qualification gives an academic adequate research training, which contributes to higher research output. This finding is consistent with Bently’s (2011) finding of a positive influence of doctoral qualifications on research output in Australia.

Another finding is that the estimated effect of the log of weekly Internet use by academics on research output is positive. The elasticity is 0.16, and the variable is statistically significant at 5 per cent level. The interpretation of the result is that a 1 per cent increase in weekly use of the Internet results in a 0.16 per cent increase in average weighted research output.

The findings reported here are important because in addition to research grants and doctoral qualifications, Internet use contributes positively to increasing research outputs. While scholars give importance to the role of the Internet in teaching and learning at an academic institution, our study has found an added importance of the Internet in academic research.

Consistent with previous research, this study has identified the positive influence of research grants on productivity as well. Therefore, any public policy directed towards limiting the availability of research funds is not conducive to enhancing research output. In the context of a competitive research grant at the national level, any public policy directed towards regional universities granting block allocation of research grants is a timely and useful option. This is a favourable policy that removes inhibiting factors in public universities for greater research output.

6. CONCLUSION AND LIMITATIONS

This study examines the relationship between Internet use by academics and their research performance in a regional university in Australia. It has used cross-section data and a simultaneous model. The main finding is that the elasticity of the effect of Internet use on academic research output is 0.16, which suggests that an increase in academics’ weekly Internet use (average) by 1 per cent (the unit of measurement is an hour) leads to an average increase in research output by 0.16 per cent, if other things remain the same. The positive relationship between Internet use and academic research performance is sufficient to argue that the benefits of the Internet are not limited to teaching, and education policy makers can contemplate the benefits in terms of increasing academic research opportunities too.

The other research findings are that the effects of research collaboration on research performance are insignificant, and that a research grant and a doctoral degree are important determinants of academics’ research output. The policy implications are: the availability of high-speed Internet must be ensured for all academics and greater investment in this sector would be a worthy public policy. Academic staff should be provided with adequate training for the proper use of Internet facilities. Further, as research grants are a positive contributor to research output, funding cuts for public universities will be counterproductive; the allocation of more research grants will be a timely and useful policy option. Finally, academics without a doctoral degree must be encouraged to become engaged in research training as soon as possible. In addition, hiring more academics with doctoral degrees will contribute to research productivity.

However, care needs to be taken when interpreting the results of this study. First, research and teaching are two outputs of academics (faculty) in Australian universities; other outputs are public service and consultancy (Dundar and Lewis, 1998). As inputs are not divisible between the research and teaching outputs, the measurement of research output is not without difficulties. ‘Evaluating research performance is an inexact science’ (Dundar and Lewis, 1998, pp. 625–26). The research productivity of an academic may vary with different measures of such productivity.

The main limitation of this study is data. It has used data that is not representative for all Australian universities, so generalisations of results of this study may not be completely appropriate. Second, this study is based on self-reported data, which, in empirical research, are not without problems as the estimated results might have some bias. Further, this study has not considered the quality of research publications. Publications in Australian Research Excellence-ranked journals and journals with high-impact factors are considered quality research, and this might be used as an indication of quality in future research. Finally, the innate ability of researchers has been overlooked in this study due to the nature of the data. Panel data and panel-data econometric modelling can handle the issue efficiently, which might also provide scope for further research. In spite of these limitations, our research findings give some indication about the benefit of Internet use, and based on this, further research is recommended using representative data and a representative sample size.

Footnotes

A Matrix of the Literature Review

Literature	Country	Research Output Measures	Level of Analysis and Data	Data Analysis Methods	Statistically Significant Determinats Found in the Studies
Ramsden (1994)	Australia	Research output index	Individual cross section	Bivariate	Research interest, involvement in research activities and academic seniority (+ve)
Dundar and Lewis (1998)	USA	Number of articles in peer-reviewed journals	Individual cross section	Standard multiple linear regression	Departmental culture such as shared attitudes (+ve) Faculty size (large or small) (+ve)
Ho (1998)	Hong Kong	Weighting and counting publication outputs	Institutional level cross-sectional	Analysis of variance (ANOVA)	University reputation (+ve) Academic rank (+ve) Experience (+ve)
Teodorescu (2000)	Australia, Brazil, Chile, Hong Kong, Israel, Japan, Korea, Mexico, the United Kingdom and the United States	Number of journal articles in the last 3-year periods	Individual cross-country	Standard linear regression for each of the 10 courtiers	Membership of professional society, attendance of conference and research grants (+ve)
Abbott and Doucouliagos (2004)	Australia	Weighted average of research publications	Institutional- level panel data, 35 government universities, 5-year period	Dynamic panel data regression	Research grant (+ve)
Lee and Bozeman (2005)	USA	Several publications in peer-reviewed journals	Survey data at the university level	2SLS test of hypothesis	Research grants (+ve), citizenships (+ve) and collaborative strategy (+ve)
Abramo et al. (2008)	Italy	Scientific publications (number) in international journals	Survey data among 78 Italian universities, individual level	Cross-tabulation analysis	Extramural research collaboration (+ve)
Mamiseishvili and Rosser (2010)	USA	Number of journal articles, conference presentations, book chapters, books and reviews in the past 2 years	Individual cross-sectional	Structural equation modelling	International faculty (+ve)
Salaran (2010)	Australia	Productivity index	Cross-sectional data at the individual level from five Australian universities	Correlation linear regression	Social interaction (+ve)
Bently (2011)	Australia	Weighted sum of self-reported publication in the 3-year period prior to the survey years	Individual cross-sectional survey data	Standard linear regression analysis	Gender (none) Academic rank (+ve) Doctorate degree (+ve) Research fund (+ve) International research collaboration (+ve)
Lissoni et al. (2011)	French and Italy	Number of articles and journal impact factor	Cross-sectional data	Generalised Tobit model	Size and nature of international collaboration (+ve), age (–ve) Gender has differential impact in France and Italy. Females publish less than males
Webber (2011)	USA	Number of peer and non-peer-reviewed articles, book reviews, book chapters and other creative works in the past 2 years	Individual cross-section data	Two-level hierarchical generalised multiple regression (Poisson regression model)	Foreign-born faculty (+ve)
Padilla-Gonzalez et al. (2011)	USA Canada Mexico	Number of articles in academic books in the last 2 years	Institutional-level cross-sectional data	Standard multiple regression	Gender gap in USA (none) Canada (+ve), Mexico (+ve)
Iqbal and Mahmood (2011)	Pakistan	No of research papers in the past in peer-reviewed journals	Individual, cross section	Cross-tabulation	Faculty teaching load (–ve)
Jung (2012)	Hong Kong	A sum of journal articles, book chapters, books and edited books over last 3 years	Individual, cross section	Standard multiple regression	Gender (+ve) Teaching experience (years) (+ve) Time on teaching (–ve) Time on research (+ve)
Meyer (2012)	USA	–	Individual	Qualitative study—interview	Mixed reaction about the influence of online teaching on research productivity
Beerkens (2013)	Australia	Number of articles Amount for research grants Weighted number of publications	Institutional-level panel data, 36 universities, 13-year period	Panel data regression	Research management (+ve)
Mishra and Smyth (2013)	Australia	Publications in law journals	Individual level, cross-sectional	Instrumental variable approach, Lewbel (2012) identification strategy	Academic rank has no effect
Fukuzawa (2014)	Japan	Number of research publications	Cross-sectional data, 39 universities in Japan	Tobit regression	Previous research experiences (+ve)

List of Relevant Questions for Participants

Notes

References

Abbott

, & Doucouliagos

(2004). Research output of Australian universities. Education Economics, 12(3), 251–65.

Abramo

, D’Angelo

C.A.

, & Di Costa

(2008). Research collaboration and productivity: Is there correlation? Higher Education, 57(2), 155–71.

Allen

G.N.

, Burk

D.L.

, & Davis

G.B.

(2006). Data collection in electronic environements: Defining acceptable use of Internet. MIS Quartelly, 30(3), 599–610.

Australian Bureau of Statistics (2008). Research and development expenditure. Canberra, ACT, Australia: The Commonwelath Government of Australia.

Beerkens

(2013). Facts and fads in academic research management: The effects of management practices on research productivity in Australia. Research Policy, 42(9), 1679–93.

Belfield

C.R.

(2000). Economic principles for education: Theory and evidence. New York: Edward Elgar.

Bently

(2011). Gender differences and factors affecting publication productivity among Australian university academics. Journal of Sociology, 48(1), 85–103.

Cameron

A.C.

, & Trivedi

P.K.

(2005). Microeconometrics: Methods and application. New York: Cambrige University Press.

Cardona

, Kretschmer

, & Strobel

(2013). ICT and productivity: Conclusions from the empirical literature. Information Economics and Policy, 25(3), 109–25.

10.

Carrington

, Coelli

, & Rao

D.S.P.

(2005). The performance of Australian universities: Conceptual issues and preliminary results. Economic Papers, 24(2), 145–63.

11.

Cohn

, & Geske

T.G.

(1992). Private nonmonetary returns to investment in higher education. In Cohn

, & Geske

T. G.

(Eds), The economics of American higher education (pp. 173–95). Netherlands: Springer.

12.

Cohn

, Rhine

S.L.

, & Santos

M.C.

(1989). Institutions of higher education as multi-product firms: Economies of scale and scope. The Review of Economics and Statistics, 71(2), 284–90.

13.

Dundar

, & Lewis

D.R.

(1998). Determinants of research productivity in higher education. Research in Higher Education, 39(6), 607–31.

14.

Edgar

, & Geare

(2011). Factors influencing university research performance. Studies in Higher Education, 38(5), 774–92.

15.

Efron

, & Tibsharani

(1993). An introduction to Bootstrap. London, UK: Champman and Hall.

16.

Fukuzawa

(2014). An empirical analysis of the relationship between individual characteristics and research productivity. Scientometrics, 99(3), 785–809.

17.

Greene

W.H.

(2012). Econometric analysis. Upper Saddle River, NJ: Prentice Hall.

18.

Gujarati

D.N.

(2003). Simultaneous-equation models. In Gujarati

D.N.

(Ed.), Basic Econometrics (pp. 103–203). New York: McGraw-Hill.

19.

Hanushek

E.A.

(1986). The economics of schooling: Production and efficiency in public schools. Journal of Economic Literature, 24(3), 1141–77.

20.

K.K.

(1998). Research output among the three faculties of business, education, humanities & social sciences in six Hong Kong universities. Higher Education, 36(2), 195–208.

21.

International Telecommunication Union (2011). Measuring the information society. Geneva, Swizerland: International Telecommunication Union.

22.

Iqbal

M.Z.

, & Mahmood

(2011). Factors related to low research productivity at higher education levels. Asian Social Science, 7(2), 1–15.

23.

Jordan

P.J.

, Chapman

R.L.

, Grimmer

, & Christie

A.M.H.

(2013). Australian and New Zealand Academy of management research producitivity survey report 2008–2010. Southport, QLD, Australia: Australian and New Zealand Academy of Management.

24.

Jovanovic

, & Rousseau

P.L.

(2005). General purpose technologies. In Aghion

& Durlauf

Steven (Eds), Handbook of economic growth (pp. 20–32). Amsterdam, Netherland: Elsevier.

25.

Jung

(2012). Faculty research productivity in Hong Kong across academic disciplines. Higher Education Studies, 2(4), 1–8.

26.

Kartz

J.S.

, & Martin

B.R.

(1997). What is research collaboration? Resesarch Policy, 16(1), 1–18.

27.

Lee

, & Bozeman

(2005). The impact of research collaboration on scientific productivity. Social Studies of Science, 35(5), 673–702.

28.

Lewis

D.R.

, & Dundar

(1995). Economies of scale and scope in Turkish universities. Education Economies, 3(2), 133–56.

29.

Lissoni

, Mairesse

, Montobbio

, & Pezzoni

(2011). Scientific productivity and academic promotion: A study on French and Italian physicists. Industrial and Corporate Change, 20(1), 253–94.

30.

Mamiseishvili

, & Rosser

V.J.

(2010). International and citizen faculty in the United States: An examination of their productivity at research universities. Research in Higher Education, 51(1), 88–107.

31.

Mamun

S.A.K.

(2012). Stochastic estimation of cost frontier: Evidence from Bangladesh. Education Economics, 20(2), 211–27.

32.

Meyer

K.A.

(2012). The influence of online teaching on faculty productivity. Innovative Higher Education, 37(1), 37–52.

33.

Mishra

, & Smyth

(2013). Are more senior academics really more research productive than junior academics? Evidence from Australian law schools. Scientometrics, 96(2), 411–25.

34.

Murphy

, Farley

, Midgley

, Carter

, Lane

, & Hafeez-Baig

(2013). An exploration of acccess to and use of mobile technologies for learning activities among higher education students. Paper presented at the research colloquia and showcase, Springfield, Australia.

35.

Okafor

E.E.

, Imhonopi

, & Urim

U.M.

(2011). Utilisation of Internet services and its impact on teaching and research outputs in private universities in South-Western Nigeria. International Journal of Emerging Technologies and Society, 9(2), 135–51.

36.

Organ

, & McGurk

(1996). Surfing the Internet and academic research: What use for Historians ? Australian Academic & Research Libraries, 27(1), 31–39.

37.

Over

(1982). Does research productivity decline with age? Higher Education, 11(5), 511–20.

38.

Padilla-Gonzalez

, Metcalfe

A.S.

, Galaz-Fontes

J.F.

, Fisher

, & Snee

(2011). Gender gaps in North American research productivity: Examining faculty publication rates in Mexico, Canada, and the US. Compare—A Journal of Comparative and International Education, 41(5), 649–68.

39.

Ramsden

(1994). Describing and explaining research producitivity. Higher Education, 28(2), 207–26.

40.

Salaran

(2010). Research productivity and social capital in Australian higher education. Higher Education Quartely, 64(2), 133–48.

41.

Seyal

A.H.

, Rahman

M.N.A.

, & Rahim

M.M.

(2002). Determinants of academic use of the Internet: A structural equation model. Behaviour & Information Technology, 21(1), 71–86.

42.

Spencer

C.E.

, & Berk

K.N.

(1981). A limited imformation specification test. Econometrica, 49(4), 1079–85.

43.

Teodorescu

(2000). Correlates of faculty publication productivity: A cross-national analysis. Higher Education, 39(2), 201–22.

44.

Zhang

(2013). Income disparity and digital divide: The internet consumption model and cross-country empirical research. Telecommunication Policy, 37(6–7), 515–29.

45.

Webber

K.L.

(2011). Faculty research productivity at US institutions: The contribution of immigration status and other selected characteristics. Paper presented at the Higher Education Institutional Research Network Conference, Kingston University, USA.

46.

Wissen

L.J.

, & Golob

T.F.

(1990). Simultaneous-equation systems involving binary choice variables. Geographical Analysis, 22(3), 224–43.

47.

Wooldrige

J.M.

(2002). Econometric analysis of cross-section and panel data. London: MIT Press.

		Sensitivity Analysis
		2SLS (1)	3SLS (2)	Bootstrap (3)	Without a Variable (4)
ln (res. output)	Age	–0.002(0.38)	–0.010(2.88)**	–0.001(0.35)	–
Panel A	ln (grant)	0.033	0.013	0.033	0.033
		(3.39)**	(1.94)	(3.79)**	(3.44)**
	ln (Internet)	0.164	0.330	0.16	0.149
		(2.12)*	(6.93)**	(2.26)*	(2.25)*
	ln (teaching)	0.003	–0.002	0.003	0.001
		(0.09)	(0.10)	(0.11)	(0.02)
	Qualification (0 = otherwise)
	1 = doctorate	0.295	0.138	0.30	0.289
		(2.27)*	(1.20)	(2.57)*	(2.27)*
	Collaboration (0 = no)
	1 = yes	0.096	0.170	0.096	0.082
		(0.50)	(1.60)	(0.61)	(0.44)
R-square		0.88	0.85	0.85	0.89
ln (Internet)	ln (res. output)	2.147	2.079	2.15	2.147
Panel B		(4.00)**	(4.31)**	(4.15)**	(4.00)**
	Age	0.034	0.033	0.034	0.034
		(3.60)**	(3.75)**	(3.61)**	(3.60)**
	Sex
	Male	0.008	–0.021	0.008	0.008
	Qualification (0 = otherwise)	(0.03)	(0.13)	(0.03)	(0.03)
	1 = doctorate	–0.341	–0.163	–0.34	–0.341
		(0.78)	(0.40)	(1.00)	(0.78)
R-square		0.91	0.90	0.90	0.90
Collaboration	ln (res. output)	0.17	0.170	0.17	0.166
		(1.59)	(1.75)	(1.26)	(1.59)
Panel C	Rank 0 = associate lecturer
	1 = lecturer	0.681	0.679	0.68	0.681
		(5.95)**	(6.47)**	(3.74)**	(5.95)**
	2 = senior lecturer	0.659	0.668	0.66	0.659
		(5.01)**	(5.55)**	(3.24)**	(5.01)**
	3 = associate professor	0.699	0.712	0.70	0.699
		(4.78)**	(5.33)**	(3.82)**	(4.78)**
	4 = professor	0.680	0.711	0.68	0.680
		(4.36)**	(4.99)**	(3.36)**	(4.36)**
	Native language 0 = otherwise
	1 = native	0.157	0.141	0.15	0.157
		(2.93)**	(2.90)**	(1.79)	(2.93)**
		0.96	0.97	0.96	0.95
R-square
Hausman test3SLS vs 2SLS	Chi-square (16) = 5.52; p-value = 0.99
N		67	67	67

Modelling a Causal Relationship between the Internet and Academic Research Performance in an Australian University: A Case Study

Abstract

Keywords

1. INTRODUCTION

2. BACKGROUND

2.1 An Overview

3. METHODOLOGY

3.1 Area of Study

4. ECONOMETRIC MODELS

Table 1 Endogeneity Test Results Variables Durbin (score) Wu–Hausman Internet x ( 1 ) 2 = 4.21; p = 0.04 F(1,63) = 4.22; p = 0.04 Collaborative research x ( 1 ) 2 = 13.86; p = 0.00 F(1,63) = 16.44; p = 0.00

6. CONCLUSION AND LIMITATIONS

Footnotes

A Matrix of the Literature Review

List of Relevant Questions for Participants

Notes

References

Table 1
Endogeneity Test Results

Variables Durbin (score) Wu–Hausman

Internet $x_{(1)}^{2}$ = 4.21; p = 0.04 F(1,63) = 4.22; p = 0.04

Collaborative research $x_{(1)}^{2}$ = 13.86; p = 0.00 F(1,63) = 16.44; p = 0.00