The impact mechanism of the controlling system in hospitals on quality of care: A study on clinical practice in China

Abstract

BACKGROUND:

Quality control system is one of the hospital information systems. The adoption of quality control system increases the work efficiency; however, to some extent, it also increases the workload for physicians.

OBJECTIVE:

The purpose of this study is to investigate the impacts of the quality control system on quality of care (e.g., process and outcome performance).

METHODS:

Our study collected physicians’ behavior information from a large urban hospital in China. We constructed the fixed-effect model to examine the relationship between the quality control system adoption and quality of care.

RESULTS:

Using the quality control system has a significant ( $p<$ 0.001) and negative effect on patients’ stay length in the hospital (process performance). Furthermore, using the quality control system has a significant ( $p<$ 0.001) and positive effect on the trends of cure rate in the hospital (outcome performance). The coefficient of the dependent variable from the patients’ stay length (process performance) is lower than the trends of cure rate (outcome performance).

CONCLUSIONS:

The controlling system can improve medical quality even though it limits physician behavior to some extent. The controlling system improves both the process performance and outcome performance, and it brings more benefits to outcome performance rather than process performance which means the reflection of the new technology may have more evident on outcome variables.

Keywords

Quality control system process quality physician’s behavior change hospital information system

1. Introduction

Work overload is typical for the top three hospitals in China [1]. To mitigate this situation, many cities invest money in the hospital information system (HIS) to improve work efficiency, such as EMR and operational system. However, in a practical application, there are still various problems mainly concentrating on the interaction between human and computer [2, 3, 4, 5].

To some extent, the investment of information systems enhanced efficiency and quality of work in hospitals [6, 7]. On the contrary, human-computer interaction also put extra pressure on agents in hospitals; for example, physicians will spend time on typing patient records especially when they have lower computer literacy. Additionally, when physicians fail to fulfill the contents of patients in HIS accurately and timely, undesirable outcomes will be accumulated in the system over time, like limitating physicians’ behavior and creating bad records. Thus, while hospital executives have to consider the effects of IT on employees, they also need to consider the differences among various systems and impacts of these tools on patients’ health status, which will potentially be very different on improving productivity, efficiency, and quality.

As a primary system in the hospital, electronic medical record (EMR) plays critical roles in improving quality of care in hospital [8, 9, 10]. However, in addition to these mainstream systems in hospitals, augmented systems are also crucial to the various organizational environment [10]. Therefore, referring to the impacts of information systems in hospitals, controversy still exists due to the complex interaction between human and diverse systems [11]. Acting as a strategy to attain a competitive advantage, quality improvement increases the reputation and profitability of a health organization during the time [12]. Hospitals have found that it is necessary to measure, monitor, and improve the quality of healthcare services to survive and achieve patient satisfaction [13, 14]. The quality control system provided a procedure which reduces errors by eliminating illegible orders, improving communication, tracking orders, checking for inappropriate prescriptions, and reminding professionals of actions to be undertaken [15, 16]. Based on the above, we pose the following question: How can quality control systems in hospitals influence the physicians’ behavior and improve the quality of care?

Previous studies have found that quality control system brings benefits to medical quality which enhances the patients’ satisfaction of medical services [30]. The benefits include reduced mortality rates [17, 18, 19], improved vaccination rates [20], increased use of recommended procedures [21], and patient safety [22, 23]. The quality control systems also have positive impacts on the ward round which has been seen to affect the quality of patient outcomes [24]. Utilizing the system to improve the ward round structure can also improve physicians’ service quality in return [25]. The system can decrease the nonstandard medical advice which results in patients’ noncompliance. This strengthens the relationship between physicians and patients and enhances patients’ compliance and outcomes [26]. To date, we still confront some research gaps. First, research related to the quality control system are still insufficient while most of them focus on the EMR, HIS and other clinical IT [9], limited studies investigated the effect of controlling systems. Second, the differences between monitoring and controlling system also raise the concerns and discussions, in particular referring to the developing country like China. Whether the quality control system relieves the stress of overworked clinicians, brings them convenient or increases the difficulty of their work are still unknown. Third, whether the system can change physicians’ behavior or not is also a novel research point.

Considering the imbalance between effective use of information system and increased workflow resulted from the information system, we leverage this gap to evaluate the impacts of the hospital information system, in particular, the effects of a quality control system on quality of care. To test this relationship, we constructed an empirical model by theorizing the relationship between the information system and behavior change. Then, we used a fixed effect model and included a control for relevant observables (e.g., workloads, environment factors) to estimate these impacts. We disentangled the variation of effects from different types of variables, in particular, the process and outcome variables. By collecting data from a large hospital in China, we found that the quality control system benefits both the process and outcome performance. However, the impacts on process variables are less significant than outcome variables. This study contributes to the literature of information system impacts in healthcare, and also promotes better understanding of economic and policy literature. In practice, our work gives support to the hospital reform related to the information system.

We organized the rest of the paper as follows. First, we review the literature of HIS and quality control systems’ impacts. Next, we introduce the research settings including a description of the empirical methods, the proposed model and the dataset. Lastly, we introduce the analytical results and then a discussion including limitations and a conclusion.

2. Methods

Process variables are variables related to medical quality, which may dynamically change before hospital discharge. Outcome variables are closely related to medical quality, which indicates the eventual consequence of disease therapy, such as improvement in function, recovery or survival. Hospital information systems provide various functions through which could improve both the effectiveness and efficiency of physicians’ decision-making [27]. To investigate the impacts of the quality control system in hospital and its various impacts on the outcome and process performance, we collected data from hospitals which have adopted the quality control system. In this section, we describe the data that we collected and discuss the operationalization of the variables that we use in our model.

2.1 Data: The setting

2.1.1 Data collection

We collected data from a large hospital in China that had applied quality control system for couples of years. It is the first time they implement such system to control medical quality. Our data collection consisted of two steps. First, the authors have several discussions with physicians and managers in this hospital. Based on these interviews, we could list several examples and the real practical problems during the process of hospital management to illustrate the impact of quality control on the behavior of agents. Second, we collected the data from three datasets including the medical record (i.e., first page of patients’ medical record, outcomes on patients’ perspective), evaluation report (i.e., comprehensive evaluation result for medical care of primary section), and department report (i.e., monthly statement of the basic section), ranging from 2015/11 to 2016/06.

2.1.2 Dependent variables

We extracted time-series data on patient admission length and defined it as a process variable, cured rate, recovery rate, mortality rate as outcome variables. We extracted this part of data from patient discharge reports based on the individual level. We aggregated these into department level by calculating the average number. The patient discharge reports include patients’ basic admission statement and time period, which supports the hypotheses of process and outcome variables as the measurements of medical quality.

2.1.3 Independent variables

We calculated the evaluation scores at the department level. As the results of system generated information, evaluation scores are the average score which measures the physician adherence behavior to the system from time and content perspectives. We extracted the scores from the quality control system which marks them automatically when users make errors. Furthermore, the system will report the final scores every month per department.

2.1.4 Control variables

We extracted the patient number, switching-out number and bed utilization rates from monthly medical report. The monthly medical report is a traditional way to measure the medical quality which reflects the medical performance from an organizational perspective. Thus, we can obtain statistic data from the control variables in this study. Additionally, we also consider the economic factors and other extrinsic factors to avoid sample bias. Table 1 presents the definition of variables.

Table 1
Definitions of constructs

Variables	Measurements	Definitions
Process performance	Length of stay	Days patient spent in the hospital.
Outcome performance	Cured rate	The percentage of cured patients when they leave the hospital.
	Recovery rate	The percentage of patients with improved condition when they leave the hospital.
	Mortality rate	The percentage of patients who died when they leave the hospital.

2.2 Model specification

We performed natural logarithm transformations on all variables to endorse the assumption of normality. Moreover, we used ordinary least square regression (OLS) to test our hypotheses, variables have introduced in the data section. However, the rate ranges from 0 to 1 which do not belong to the normal distribution and is inconsistent with the basic assumption of OLS, so we used a fractional logit regression model to examine the impacts on dependent variables related to rate. Furthermore, to examine how the quality control system affects the quality of care, we use panel data and a fixed effect regression model as follows:

$\displaystyle\textit{ServiceOutcome}_{it}=A_{s}+b*\text{Log}(\textit{% ServiceQuanlity}_{it})+d*C_{st}+e_{it}$

Here, $\textit{ServiceOutcome}_{it}$ is the patient outcome variable for a department $i$ at time $t$ which indicates the discharge status when they leave the hospital for many reasons. $\textit{ServiceQuanlity}_{it}$ captures the average scores of department $i$ at time $t$ . $A_{s}$ refers to the department fixed effect. And $C_{st}$ is the department control variables.

In the above specification, the department level fixed effect controls the time-invariant differences across departments. The inclusion of this fixed effect compares each department in a given month to any other department at other periods. Additionally, other factors are influencing the time-serious variation of impacts; in other words, the fixed effect cannot be the only measurement. To account for such effects, several control variables $C_{st}$ are included in the model specification to account for factors that vary within each department over time. Details of these control variables have been delineated earlier in the data section. For all model specifications, we clustered the error terms at the department level to account for autocorrelation in the data [28]. Observations are at the department-month level based on the monthly medical reports across all in-sample patients admitted to this hospital. Accordingly, there are 207 observations in total and united in balance panel data for 23 departments of 9 periods.

3. Results

3.1 Main results

We have two dependent variables including procedural measurements and outcome measurements. Table 2 presents summary statistics of variables. Table 3 shows the results for the fixed effect model.

Table 2
Summary statistics

Variables	Obs.	Mean		Std. dev.		Min.		Max.
Department number ${}^{*}$	207	12		6	.65	1		23
Evaluation score	207	870	.42	88	.11	592		1156	.95
Log average length of stay	207	10	.90	4	.23	2	.07	32	.92
Mortality rate	207	0	.02	0	.02	0		0	.12
Cured rate	207	0	.35	0	.34	0		1
Recovery rate	207	0	.95	0	.06	0	.75	1
Initial patient numbers	207	37	.95	28	.53	0		142
Switching-in rate	207	0	.028	0	.034	0		0	.17
Readmission rate	207	0	.92	0	.14	0	.58	1	.52
Undiagnosed rate	207	0	.05	0	.06	0		0	.27
Diagnose uncertainty rate	207	0	.01	0	.03	0		0	.25
Switching-out rate	207	0	.03	0	.03	0		0	.15
Bed utilization rate	207	90	.08	27	.25	4		154	.19

Table 3

Results of the model

Variable	Dependent variable		Dependent variable
	Process: Length of stay		Outcome: Cure rate
Evaluation score	$-$ 0.787 ${}^{***}$	(0.189)	2.023 ${}^{**}$	(1.208)
Initial patient numbers	$-$ 0.043	(0.078)	$-$ 0.421 ${}^{***}$	(0.143)
Bed utilization rate	$-$ 0.002 ${}^{**}$	(0.001)	$-$ 0.022 ${}^{***}$	(0.005)
Readmission rate	0.340 ${}^{**}$	(0.144)	1.346	(0.889)
Undiagnosed rate	0.433	(0.334)	$-$ 7.163 ${}^{***}$	(2.466)
Diagnose uncertainty rate	0.327	(0.602)	$-$ 2.093	(3.061)
Switching-out rate	$-$ 0.390	(0.542)	$-$ 5.336	(3.943)
Switching-in rate	1.170 ${}^{**}$	(0.579)	$-$ 1.033	(4.291)
Intercept			$-$ 11.626	(8.091)
R-squared	0.183		0.383
Sample number	Balanced panel: $n=$ 23, $T=$ 9, $N=$ 207

3.1.1 Process performance

The results of the procedural measurements model suggest that quality control system has a significantly negative effect on the procedural variables ( $p<$ 0.001). Since we use the average length of stay to measure the process performance, the results are negative and consistent with models which indicates the decrease of stay length in hospital.

3.1.2 Outcome performance

The results of the outcome measurements model suggest that quality control system has a significantly positive effect on the procedural measurements ( $p<$ 0.001). Since we use the cure rate to measure the impact of the system, the results are positive and consistent with models which indicates the increasing trends of cure rate in hospital.

3.1.3 Comparison

According to the results from Table 3, the coefficient of the dependent variable from the process is lower than the outcome, and they are all significant.

3.2 Post hoc analysis

After we test the main impacts of the information system, we identify the variation of scores on departments. According to Fig. 1, the average scores present upward trend in the period of system implementation. Limited by the data we collected, the trend before the time point, 2015/11, is stable but not visible. But it is obvious that increasing trend gets to the highest point at 2015/11 and then comes to considerable downward until 2015/12. Comparing with increase occurred before 2015/11, the substantial increase occurred from 2015/12 to 2016/1 with a gradually steady trend. According to this fluctuation, the marked downward and then upward show the reflection related to the first implementation of a new information system which may occur because of the sudden limitation and the inconsistent with traditional medical process. Furthermore, a stable trend reveals the forming of new habits of the medical process which requires further exploration.

Figure 1.

Trends of evaluation scores of departments.

3.3 Robustness checks

In this section, we assess the robustness of the main results concerning alternative variables of dependent variables. Since our model formulation is at the department level and the patient status can be different in different departments and situations, we replaced dependent variables to make sure that our analysis is robust. We use these two-replaced outcome dependent variables, recovery rate and mortality rate from our data, in that these are also two dependent variables which can reflect the patient status when they leave the hospital. Table 4 presents the robustness results in which are all significant and consistent with results before. In other words, according to the robustness test, the impacts on outcome variables are still more significant than process variables during the medical process.

Table 4
Results of robustness checks

Variable	Dependent variable
	Recovery rate		Mortality rate
Evaluation score	2.290 ${}^{***}$	(0.667)	$-$ 2.534 ${}^{**}$	(1.025)
Initial patient numbers	$-$ 0.146	(0.114)	0.245	(0.152)
Bed utilization rate	$-$ 0.000	(0.004)	0.020 ${}^{***}$	(0.006)
Readmission rate	0.814	(0.666)	$-$ 1.357	(0.960)
Undiagnosed rate	1.142	(1.351)	$-$ 0.717	(1.674)
Diagnose uncertainty rate	$-$ 2.004	(2.415)	$-$ 1.361	(3.139)
Switching-out rate	1.104	(3.596)	8.664 ${}^{*}$	(5.166)
Switching-in rate	$-$ 1.177	(3.194)	0.794	(4.813)
Intercept	$-$ 12.827 ${}^{***}$	(4.377)	11.120	(63987)
R-squared	0.05		0.121
Sample number	Balanced panel: $n=$ 23, $T=$ 9, $N=$ 207

3.4 Key findings

According to the results of models, our study has the following findings. First, the controlling system can improve medical quality even though it limits physician behavior. Previous research showed the potential of the monitoring system in the hospital rather than controlling system because of the resistant emotion towards new technology and some habits [23]. Contrarily, our study gives strong support to the benefits of controlling system which provides new insight for both theoretical and practical perspectives. Moreover, it is an inventive way for physicians to improve the adherence of the system. Second, the controlling system improves both process quality and outcome quality. Previous research explored the effect on quality of care [29], as well as the effects of surveillance, prevention, and controlling system [30]. Our study puts forward a new insight to measure medical quality from process and outcome perspective. Third, comparing with outcome quality, the impacts of controlling system on process quality has a less significant effect which means that the reflection of the new technology may be more evident on outcome variables. When we discuss process quality, it is sometimes a two-sided sword which is sensitive to the change. The outcome variables may have more direct effects from variations about the work process. We discuss the theoretical and practical implications of the study in the following section.

3.5 Main implications

First, we extend the research about the impacts of different types of subsystems in hospitals, in particular, the role of the quality control system. Extant research on the controlling system and its effects on quality of care state that monitoring the activities of clinicians in hospitals is more valuable than controls [23]. Our findings reveal the positive impacts of the quality controlling system which tests the roles in turn and posits new research about different types of information system in hospitals. Additionally, our findings examined the various impacts on medical performance, such as process variables and outcome variables. By variable classification, our findings answer the questions about the mechanism of effectiveness and efficiency related to the clinical information system and extend the exploration of the process to outcome. Specific classifications will help us better understand the impact and controversy results from previous research.

Second, our findings make it clear that the implementation of the controlling system will benefit organizational performance even though it limited and changed the behavior of physicians. Thus, considering the low costs of this subsystem, hospitals can consider to adopt it as a way to mitigate the drawbacks from other influences.

4. Conclusions

In conclusion, our study provides new insights into the roles of quality control functions on two main types of quality care, namely process and outcome performance. Our findings related to these two variables indicate that controlling system promotes quality of care by regulating user behavior. Even though such a system limits physician behavior, and increases the workflow to some extent, the results still show the positive side which argues that the controlling system is not useless comparing with a monitoring system in the hospital. We believe that this research can be extended to explore a variety of technology and policy interventions related to medical process such as enacting personalized system design for specific departments to satisfy their requirements.

Furthermore, since this study used data from a department level which is still from the organizational perspective, further research could be done from the individual level to explore the variation of behavior change, such as agents’ adherence with the various controlling system specifications, and also interactions among different types of system in the hospital. In particular, this research can serve as a basis for further studies on how behavior intervention in quality management can be combined with technological mechanisms to produce optimal quality outcomes. We hope that our results will spawn further research in this critical area.

Footnotes

Acknowledgments

This work received support from the National Natural Science Foundation of China (Grant nos 71531007, 71471048, and 71471049).

Conflict of interest

None to report.

References

Dai

Wang

Ayala

. Medical informatics and the “three long, one short” problem of large urban hospitals in China. JAMA. 2016; 316(3): 269-70. doi: 10.1001/jama.2016.5487.

Carneiro

Pimenta

Gonçalves

Neves

Novais

. Monitoring and improving performance in human-computer interaction. Concurrency and Computation: Practice and Experience. 2016; 28(4): 1291-309. doi: 10.1002/cpe.3635.

Hibbeln

Jenkins

Schneider

Valacich

Weinmann

. How is your user feeling? Inferring emotion through human-computer interaction devices. MIS Quarterly. 2017; 41(1): 1.

Carvalho

de Castro Andrade

de Oliveira

de Sousa Santos

Bezerra

CIM

. Quality characteristics and measures for human-computer interaction evaluation in ubiquitous systems. Software Quality Journal. 2016: 1-53. doi: 10.1007/s11219-016-9320-z.

Cairns

Bond

Finlay

Breen

Guldenring

Gaffney

, et al. A computer-human interaction model to improve the diagnostic accuracy and clinical decision-making during 12-lead electrocardiogram interpretation. Journal of Biomedical Informatics. 2016 12; 64: 93-107. doi: 10.1016/j.jbi.2016.09.016.

Ancarani

Di Mauro

Gitto

Mancuso

Ayach

. Technology acquisition and efficiency in Dubai hospitals. Technological Forecasting and Social Change. 2016 12; 113, Part B: 475-85. doi: 10.1016/j.techfore.2016.07.010.

Zakaria

Mohd Yusof

. Understanding technology and people issues in hospital information system (HIS) adoption: case study of a tertiary hospital in malaysia. Journal of Infection and Public Health. 2016 11; 9(6): 774-80. doi: 10.1016/j.jiph.2016.08.017.

Dranove

Garthwaite

Ody

. Investment subsidies and the adoption of electronic medical records in hospitals. Journal of Health Economics. 2015; 44(9): 309-19.

Xiao

Sharman

Singh

Danzo

Rao

. “Meaningful Use” of ambulatory EMR: does it improve the quality and efficiency of health care? Health Policy & Technology. 2012; 1(1): 28-34.

10.

Almazán

Blas-Machado

Kocan

Yoshioka

Blouin

Mangold

, et al. The impact of health information technology bundles on performance: an econometric study. Journal of Operations Management. 2016; 41(1): 25-41.

11.

Leila

. The effects of health information technology on the costs and quality of medical care. Journal of Health Economics. 2014; 34(1): 19-30.

12.

Emmert

Taheri-Zadeh

Kolb

Sander

. Public reporting of hospital quality shows inconsistent ranking results. Health Policy. 2017 1; 121(1): 17-26. doi: 10.1016/j.healthpol.2016.11.004.

13.

Williams

Asi

Raffenaud

Bagwell

Zeini

. The effect of information technology on hospital performance. Health Care Management Science. 2016 2016; 19(4): 338-46. doi: 10.1007/s10729-015-9329-z.

14.

Graham

. Defining and measuring patient satisfaction. The Journal of Hand Surgery. 2016 9; 41(9): 929-31. doi: 10.1016/j.jhsa.2016.07.109.

15.

Bates

Pappius

Kuperman

Sittig

Burstin

Fairchild

, et al. Using information systems to measure and improve quality. International Journal of Medical Informatics. 1999 2/1; 53(2-3): 115-24. doi: 10.1016/S1386-5056(98)00152-X.

16.

van Wijk

van der Lei

Mosseveld

Bohnen

van Bemmel

. Assessment of decision support for blood test ordering in primary care. A randomized trial. Ann Intern Med. 2001 2001/2; 134(4): 274-81. PMID: 11182837. doi: 10.7326/0003-4819-134-4-200102200-00010.

17.

Devaraj

Kohli

. Information technology payoff in the health-care industry: a longitudinal study. Journal of Management Information Systems. 2000; 16(4): 41-68.

18.

Devaraj

Rajiv

. Performance impacts of information technology: is actual usage the missing link? Management Science. 2003; 49(3): 273-89.

19.

Amarasingham

Plantinga

Diener-West

Gaskin

Powe

. Clinical information technologies and inpatient outcomes: a multiple hospital study. Archives of Internal Medicine. 2009 Jan 26; 169(2): 108-14. PMID: 19171805. doi: 10.1001/archinternmed.2008.520.

20.

Dexter

Perkins

Maharry

Jones

McDonald

. Inpatient computer-based standing orders vs. physician reminders to increase influenza and pneumococcal vaccination rates: a randomized trial. JAMA. 2004; 292(19): 2366-71. doi: 10.1001/jama.292.19.2366.

21.

Kucher

Koo

Quiroz

Cooper

Paterno

Soukonnikov

, et al. Electronic alerts to prevent venous thromboembolism among hospitalized patients. The New England Journal of Medicine. 2005 Mar 10; 352(10): 969-77. PMID: 15758007. doi: 10.1056/NEJMoa041533.

22.

Parente

McCullough

. Health information technology and patient safety: evidence from panel data. Health Affairs (Project Hope). 2009 Mar-Apr; 28(2): 357-60. PMID: 19275990. doi: 10.1377/hlthaff.28.2.357.

23.

Aron

Dutta

Janakiraman

Pathak

. The impact of automation of systems on medical errors: evidence from field research. Information Systems Research. 2011; 22(3): 429-46.

24.

Pucher

Aggarwal

Darzi

. Surgical ward round quality and impact on variable patient outcomes. Journal of the American College of Surgeons. 2013 9; 217(3, Supplement): S103. doi: 10.1016/j.jamcollsurg.2013.07.235.

25.

Pucher

Aggarwal

Singh

Srisatkunam

Twaij

Darzi

. Ward simulation to improve surgical ward round performance: a randomized controlled trial of a simulation-based curriculum. Ann Surg. 2014 2014/08; 260(2): 236-43. PMID: 24646529. doi: 10.1097/sla.0000000000000557.

26.

Memon

Shaikh

Soomro

Shaikh

Khwaja

. Non-compliance to doctors’ advice among patients suffering from various diseases: patients’ perspectives: a neglected issue. Journal of Medicine (Bangladesh). 2017; 18(1): 10-4. doi: 10.3329/jom.v18i1.31170.

27.

Angst

Devaraj

D’Arcy

. Dual role of it-assisted communication in patient care: a validated structure-process-outcome framework. Journal of Management Information Systems. 2012; 29(2): 257-92.

28.

Bertrand

Duflo

Mullainathan

. How much should we trust differences-in-differences estimates? Quarterly Journal of Economics. 2004; 119(1): 249-75.

29.

Miller

Tucker

. Privacy protection and technology diffusion: the case of electronic medical records. Management Science. 2009; 55(7): 1077-93.

30.

Atreja

Steven

Daniel

Russell

Patrick

Committee

HICPA

. Opportunities and challenges in utilizing electronic health records for infection surveillance, prevention, and control. American Journal of Infection Control. 2008; 36(3): S37-S46.

The impact mechanism of the controlling system in hospitals on quality of care: A study on clinical practice in China

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSIONS:

Keywords

1. Introduction

2. Methods

2.1 Data: The setting

2.1.1 Data collection

2.1.2 Dependent variables

2.1.3 Independent variables

2.1.4 Control variables

Table 1 Definitions of constructs

3. Results

3.1 Main results

Table 2 Summary statistics

3.1.2 Outcome performance

3.1.3 Comparison

3.2 Post hoc analysis

Table 4 Results of robustness checks

3.5 Main implications

4. Conclusions

Footnotes

Acknowledgments

Conflict of interest

References

Table 1
Definitions of constructs

Table 2
Summary statistics

Table 4
Results of robustness checks