The moderating effect of political connection on the relationship between non-audit services and accounting quality: Evidence from an emerging market

Abstract

This study investigates whether the provision of non-audit services (NAS) is associated with accounting quality and whether political connection moderates the relationship between the NAS and accounting quality. The sample includes 2,245 firm-year observations from Malaysia during the period 2013–2017. This study uses the absolute value of discretionary accruals to measure accounting quality. We find that provision of high NAS leads to low accounting quality, exhibited by high discretionary accruals. We also find that politically connected (PCON) firms show higher discretionary accruals than non-PCON firms. However, the PCON firms weaken the positive relationship between NAS on discretionary accruals suggesting a substitutive role of political connections in determining the effect of NAS on accounting quality. We further document a non-linear association between NAS and discretionary accruals, whereby the provision of NAS would lead to low discretionary accruals after an optimal point. Therefore, regulators need to weigh the costs and benefits of regulating NAS and consider the moderating effects of PCON firms and the non-linear effect of NAS. This study adds to the limited, albeit substantial, evidence on the joint effect of PCON firms and NAS on accounting quality.

Keywords:

• non-audit services
• accounting quality
• discretionary accruals
• political connection

JEL CLASSIFICATION:

• M41
• M42
• G18

1. Introduction

Firms can enjoy significant advantages from political connections. For example, priority for government project selection (Chung et al., 2019; Sun et al., 2016), market protection (Faccio, 2010), lower insurance rights, smooth access to donations and government grants (Ben Rejeb Attia et al., 2016), bank loans privilege (C. M. Chen et al., 2013; Yang et al., 2012), preferential access to credit (Claessens et al., 2008; Kallias et al., 2022), and tax reduction (Brown et al., 2015; Y. Chen et al., 2020); are among the benefits given to politically connected firms.

A prior study on firms’ political connections reported Malaysia as the second country with the highest percentage of politically connected firms after the UK (Faccio, 2006). In a more recent study, Kamarudin et al. (2021) found varied effects between types of politically connected (PCON) firms in Malaysia—connected through business, board of directors, or family members on analysis forecast. These studies show that appointing a politician or politically linked figure in the boardroom has become dominant among firms in the country and is still persistent.

Political connectedness is rising due to the immense benefit gained from those connections and aligned with the resource dependency theory. The theory claims that the directors are essential resources to the firms, and appointing politicians in the boardroom indirectly benefits the firms, especially in government project selection (Chung et al., 2019; Sun et al., 2016) and market protection (Faccio, 2010). In a positive outlook, Masud et al. (2019) found that politically connected firms have better disclosure due to the potent networking with the government in maintaining their reputation and social status.

Although the firms benefit from that connection, there is much debate concerning the adverse effects of political links in the accounting and business fields. Empirical evidence has found that political connectedness may influence management decisions (Masud et al., 2019) and firms’ investment decisions (Bae et al., 2021). Hence, political connections can divert the firms from their fundamental interest in maximising shareholder wealth to the political interest and create agency problems (Azmi et al., 2020). Using samples across 19 countries-level, Chaney et al. (2011) found that political connections negatively affect accounting quality. Then, Khelil et al. (2021) found that political connections increase the likelihood of receiving a favourable audit opinion and are associated with higher audit fees in Asian countries. This issue undeceives the stakeholders, especially the regulators, on the politician’s involvement in the boardroom. Consequently, in 2021, the Malaysian Code of Corporate Governance was revised to limit the appointment of active politicians in the boardroom to reduce the negative effect. The revised MCCG 2021 shows that the issues of political connection in the field are persistent and pertinent.

Considering the detrimental effects of political intervention in the business, the roles of the external auditors as the external watchdog of the financial statement become a matter. Credible, objective and professional scepticism of external auditors can limit the negative effect of political connections and accounting irregularities in the firms. However, the economic bonding between audit firms and their clients may impair auditors’ independence in providing high-quality financial reporting (Beardsley et al., 2021; Kim et al., 2006; Simunic, 1984). Researchers continuously raise the issue that purchasing non-audit services (NAS) threatens auditor independence (Beardsley et al., 2021). Specifically, Abdul Wahab et al. (2015) found that PCON firms purchased higher NAS than their counterparts. The political influences and higher NAS purchased by auditee can distract audit function (Beardsley et al., 2021), impair auditor independence (Bloomfield & Shackman, 2008), and influence financial reporting quality (Abdul Wahab et al., 2020; Chaney et al., 2011). The study on NAS is relevant because Huang and Kang (2021) found that highly reputable US firms still pay higher NAS fees.

Recent studies in Malaysia by Abdul Wahab et al. (2020) and Nik Abdul Majid et al. (2021) have found that NAS negatively affects accounting quality. Abdul Wahab et al. (2020) examine NAS by classifying NAS into recurring and non-recurring NAS, and they found that both NAS negatively affect accrual quality. While Nik Abdul Majid et al. (2021) examined NAS and accounting quality by taking into account audit committee characteristics and found NAS impairs accounting quality, and the existence of audit committees moderates the negative association between NAS and accrual quality. Although they found that higher NAS compromise the accounting quality, limited evidence prevails on the effect of institutional setting such as country and regulation that are deemed necessary in examining the impact of NAS and accounting quality.

The unique institutional setting in Malaysia, where most businesses are subject to government interventions such as political influence (Abdul Wahab et al., 2009; Al-Dhamari & Ku Ismail, 2015; Sani et al., 2020), could provide different views on the issue of NAS and accounting quality. For example, Gul (2006) supported this finding that PCON firms pay higher audit fees than their counterpart. Specifically, Abdul Wahab et al. (2015) found that PCON firms purchased higher NAS than non-PCON firms. Additionally, Alareeni (2019) discovered that institutional settings such as the legal system and country could moderate the association between NAS and audit quality. Hence, political connections could influence the association between NAS and accounting quality in Malaysia.

Using a Malaysian sample of 2,245 firm-year observations from 2013–2017, we investigate and find that the PCON firms and firms that purchased the NAS are associated with low accounting quality, shown by high discretionary accruals. Furthermore, our analysis of the combined effect of NAS and PCON shows that PCON firms weaken the positive relationship between NAS on discretionary accruals, supporting the notion that PCON firms play a vital substitute role in determining the effect of NAS on discretionary accruals. Our results are robust to a battery of tests including (1) alternative measurements for discretionary accruals using the modified Jones performance-adjusted model Dechow et al. (1995) estimated based on Fama and French’s (Fama & French, 1997) 10 industry groups; (2) alternative measures for NAS using the natural logarithm for NAS and employing the number of politicians on board as an alternative proxy for PCON. We also found that the results hold even after addressing the endogeneity issue using the propensity score matching (PSM) technique. Finally, we investigate the potential of a non-linear association between NAS and discretionary accruals. We find evidence of a curve relationship where the positive effect of NAS on discretionary accruals is before it reaches the optimal point, which would be negative after the NAS optimal point.

Our results contribute to the stream of literature that examines the effects of political connections and non-audit services on accounting quality. This study differs from a study of Malaysian firms by Abdul Wahab et al. (2020), which examines the sole effects of recurring and non-recurring NAS on accrual quality without considering the impact of other variables, which could affect the relationship. While Nik Abdul Majid et al. (2021) only consider the audit committee’s characteristics as a moderating variable between NAS and accrual quality, the effect of political link is not considered. Since Abdul Wahab et al. (2015) shows that political connections are dominant determinants of audit quality of Malaysian firm, there is a need to have political connections in the picture. The literature gap provided an opportunity to further investigate this issue by examining the effect of political connections on those associations. We also contribute to the practicality in which this study aligns with the latest Malaysian Code of Corporate Governance 2021 (MCCG 2021), which restricts active politicians involved in the business (Securities Commission Malaysia, 2021). Due to the importance and significance of the Malaysian environment, this study added to the nascent research by examining the effect of NAS and political connections on the accounting quality in Malaysia.

The remainder of this paper is organised as follows: Section 2 reviews prior literature and hypotheses development. Section 3 explains and illustrates the research design. Section 4 presents the descriptive and empirical studies, and Section 5 reports our conclusions.

2. Literature review and hypotheses development

Non-audit services are the professional services provided by a qualified public accountant during the period of audit engagement that is not related to audit or review of financial statements. However, the provision of NAS charged by auditors has been debated extensively by scholars (Abdul Wahab et al., 2020; Alareeni, 2019; Bell et al., 2015; Nik Abdul Majid et al., 2021; Onulaka et al., 2019; Simunic, 1984). Section 101 of the Sarbanes-Oxley Act restricted external auditors from providing nine specified NAS to their clients and required audit committee approval to preserve auditor independence. This restriction is consistent with Malaysia’s practices that require all publicly listed firms to disclose the amount of NAS in their annual report (Bursa Malaysia, 2013); and demand audit committee and board of directors’ approval before NAS engagement (MCCG, 2017). This action is to protect shareholders’ interests and increase transparency. Although SOX 2002 limits NAS offered by auditors to their clients, NAS remains part of the revenue and economic necessity for the survival of accounting firms (Onulaka et al., 2019).

The competition in the audit market among accounting firms drives down audit fees and allows them to offer NAS. Auditing literature found that NAS purchases by auditees had reflected the economic bonding and knowledge spillover between the auditor and auditee (Parkash & Venable, 1993). Although NAS could improve knowledge spillover, it can also impair auditor independence (Bloomfield & Shackman, 2008), reduce audit quality (Beardsley et al., 2021; Bell et al., 2015; Carmona et al., 2021; Lennox, 1999), increase earnings restatement (Bloomfield & Shackman, 2008), and lessen accrual quality (Abdul Wahab et al., 2020; Nik Abdul Majid et al., 2021). In addition, prior studies have found that institutional settings such as political intervention in the boardroom further inclined agency problems and influenced the accrual quality (Al-Dhamari & Ku Ismail, 2015; Sani et al., 2020). These findings have shown that NAS can impair auditor independence, and this problem becomes serious in the presence of politicians in the firms.

Therefore, further research is still warranted to examine the effect of political connections on NAS and accounting quality.

2.1. Non-audit fees and accounting quality

Conflicting views existed on the effect of the NAS in the accounting and auditing field (Beardsley et al., 2021; Francis, 2004; Walker & Hay, 2013). First is the negative view of NAS, where NAS charged by auditors has created an agency problem and increased information asymmetry between auditors and the public. Other scholars debated that higher purchases of NAS could inflame audit quality (Alareeni, 2019) and affect financial reporting quality (Becker et al., 1998). This view proposed a negative association between NAS and accounting quality. On the other hand, the second is the optimistic view of NAS, where the purchase of NAS can encourage knowledge spillover and increased audit efficiency (Parkash & Venable, 1993; Walker & Hay, 2013). The second view proposed a positive association between NAS and accounting quality.

The empirical evidence on the relationship between NAS and accounting quality is mixed. In the first view, other scholars have found that the provision of NAS impairs auditor independence (Beardsley et al., 2021; Frankel et al., 2002). Then, auditor independence would doubt stakeholders’ financial reporting quality (Abdul Wahab et al., 2020). Using interviews among audit partners and pension fund managers, Onulaka et al. (2019) found that NAS provision impeded auditors’ independence and increased information asymmetry between the auditor and public expectation. Further, Beardsley et al. (2021) examined the distraction effect of NAS and found that minimal purchase of NAS would also distract audit function and earnings misstatement. Regardless of using qualitative and quantitative studies, researchers in auditing have found similar findings that higher or lower purchases of NAS impaired auditor independence, thus diminishing audit function in providing high-quality financial reporting.

Despite considerable research on NAS and accounting quality, researchers still lack consensus. Past studies provide a different view on the effect of NAS and found that there is limited evidence to support that higher NAS purchased by an auditee violates auditor independence (Ashbaugh et al., 2003) and increased earnings restatement (Bloomfield & Shackman, 2008). In line with Paterson and Valencia (2011) provides evidence that NAS positively affects financial reporting quality. A recent study by Carmona et al. (2021) also discovered that high NAS does not necessarily result in poor quality financial reporting, thereby casting doubt on the public perception of NAS and auditor independence. These findings support the view of economic bonding and knowledge spillover between the auditor and auditee (Parkash & Venable, 1993; Simunic, 1984). Their outcomes prove that the purchase of NAS does not necessarily impair auditor independence and accounting quality.

From the arguments above, we postulate that the provision of NAS creates economic bonding between the auditor and auditee because the auditor becomes part of the firm’s administration and decision-makers. Thus, this may impair auditor objectivity, professional scepticism, and independence in producing high-quality financial reporting (Beardsley et al., 2021). Therefore, we expected that higher NAS purchased by clients impair accounting quality, and we proposed the first hypothesis as follows:

H₁: Non-audit services purchased by firms are negatively associated with accounting quality.

2.2. Political connection, non-audit fees, and accounting quality

Scholars in auditing found mixed findings between NAS and accounting quality. Empirical studies documented evidence that higher purchases of NAS create agency problems, impair auditor independence (Onulaka et al., 2019), reduce audit quality (Alareeni, 2019; Carmona et al., 2021), and diminish accrual quality (Abdul Wahab et al., 2020; Nik Abdul Majid et al., 2021). Therefore, the auditee was expected to limit NAS purchases from their auditor, although NAS can improve knowledge spillover and efficiency (Parkash & Venable, 1993; Walker & Hay, 2013). In addition, the involvement of politicians in the boardroom as decision-makers may influence auditors’ choice and purchase of NAS. Supported by Malek & Che-Ahmad (2013) found that the director-auditor link positively influences the purchase of NAS. Moreover, Malaysia’s practices require director and audit committee approval before purchasing NAS (MCCG, 2017). Therefore, institutional settings such as political connections might influence the purchase of NAS.

Extensive studies on political connections in Malaysia show a significant portion of publicly listed firms are linked to politicians through the appointment of directors on board (Al-Dhamari & Ku Ismail, 2015; Bliss & Gul, 2012; Fraser et al., 2006; Gomez, 2014; Gomez & Jomo, 1997; Johnson & Mitton, 2003; Kamarudin et al., 2021; Wong & Hooy, 2020). Prior studies found evidence that firms with connections to politicians have better odds of increasing their share prices by up to 4% than non-PCON firms (C. M. Chen et al., 2013). This increase in share prices among PCON firms is motivated by the gained benefits via government project selection (Chung et al., 2019; Sun et al., 2016), market protection (Faccio, 2010), lower insurance rights, lower tax, and ready access to donations and government grants (Ben Rejeb Attia et al., 2016). PCON firms also benefit from state access and privilege of bank loans (C. M. Chen et al., 2013; Yang et al., 2012), preferential access to credit, and tax reduction (Y. Chen et al., 2020). In addition, shareholders in PCON firms enjoy tax savings with lower effective tax rates (ETR) in China (Y. Chen et al., 2020). Additionally, Azmi et al. (2020) and Marzuki and Shukri (2019) have found that political intervention positively affects future performance among Malaysian firms. Masud et al. (2019) also discovered that high politicians in the boardroom result in better corporate corruption disclosure.

The immense advantage that comes with political connections encouraged firms to appoint politicians in the firms. PCON firms are showered with benefits from those connections, a luxury missing in non-PCON firms. Therefore, firms prefer to appoint politicians or bureaucrats in the boardroom rather than financial experts (Fan et al., 2007). The surge of political involvement in the boardroom creates agency problems that can influence the director’s independence and divert the firm’s fundamental objective to a politician’s motive (Hasnan & Marzuki, 2017). This reality triggers serious complications because politicians in the boardroom can influence management’s decisions (Amara & Khlif, 2020; Masud et al., 2019).

Although the politician’s appointment as the director in the boardroom aligns with the resource dependency theory (Chung et al., 2019), this appointment can further increase agency problems and information asymmetry. Nonetheless, prior studies found that political connection negatively links to audit fees (E. A Abdul Wahab et al., 2011; Tee, 2018), quality of governance (Dicko, 2017), earnings quality (Harymawan & Nowland, 2016; Hashmi et al., 2018), firm performance and value (Berkman & Galpoththage, 2016; Wong & Hooy, 2020), audit quality (Khelil et al., 2021), and analyst forecast accuracy (Kamarudin et al., 2021).

Gul (2006) finds a significant increase in audit fees in PCON firms than in non-PCON firms, probably due to the high-risk exposure and worsened accounting information in political connection firms among 19 countries (Chaney et al., 2011). Similarly, E. A Abdul Wahab et al. (2011) find higher audit fees (stationary and non-stationary) for PCON firms in Malaysia. In addition, a current study by Amara and Khlif (2020) finds that PCON firms exercised their discretion to reduce accounting quality through discretionary accrual. To conceal these activities, Harymawan (2020) has found evidence that PCON firms were less likely to appoint Big 4 firms as their auditor before and after the political party reform. These findings indicate evidence that political connection influences auditors’ choice and purchases of NAS in the firms. Thus, it may influence the auditor’s objectivity, independence, and accounting quality.

Prior studies in auditing examine the direct effect of political connection, NAS, and accounting quality. For example, Paterson and Valencia (2011) investigate the effect of NAS and auditor independence, while Abdul Wahab et al. (2015) examine the effect of political connection on auditor independence. In addition, the current study by Harymawan (2020) examines the effect of political connections on auditor choice. However, only Nik Abdul Majid et al. (2021) consider the moderating effect of the audit committees on NAS and accrual quality. Hence, the lack of research examining the moderating effect of political connection on NAS and accounting quality motivated us to examine this effect.

Therefore, to better understand the effect of political connections on the effectiveness of auditors with non-audit services, this study specifically examines the moderating effect of political connections on NAS and accounting quality. We believe the political connection is necessary to explain the relationship between NAS and accounting quality. Based on the argument from prior studies and the revised MCCG 2021 that restricted active politicians’ involvement in the business, we expect political connection may strengthen or weaken the relationship between NAS and accounting quality. Hence, we proposed the second hypothesis as follows:

H₂: Political connection moderates the association between NAS and accounting quality

3. Data and research methodology

3.1. Data and sample

Our sample covers Malaysian listed firms for the period 2013–2017. We collected the financial data from the Refinitiv database, while data on NAS and politicians on boards were hand-collected from the corporate annual reports. We removed observations with missing or incomplete. We also winsorised observations that fell in the top and bottom 1% to mitigate the influence of outliers. Table 1 reports the breakdown of the final sample based on year and industry.

Table 1. Sample breakdown

Panel A: Breakdown by industry
Industry	N	Percent
Communication Services	55	2.45
Consumer Discretionary	317	14.12
Consumer Staples	363	16.17
Energy	91	4.05
Financials	5	0.22
Health Care	39	1.74
Industrials	650	28.95
Information Technology	163	7.26
Materials	397	17.68
Real Estate	159	7.08
Utilities	6	0.27
Total	2,245	100
Panel B: Breakdown by year
Year	Freq.	Percent
2013	448	19.96
2014	454	20.22
2015	448	19.96
2016	451	20.09
2017	444	19.78
Total	2,245	100

From the table, the highest number of observations comes from the industrials sector (n = 650, 28.95 percent), followed by consumer materials (n = 397, 17.68 percent), consumer discretionary (n = 317, 14.12 percent), and staples (n = 363, 16.17 percent). In terms of breakdown by year, the observations were equally distributed, ranging from 444 to 451 observations per year.

3.2. Measurement

For the dependent variable, accounting quality, we employed two measurements. First, based on Kothari et al. (2005), we used the performance-adjusted Jones model to calculate the discretionary accruals |DACC1|. Second, following Dechow et al. (1995) , we applied the modified Jones model to estimate the discretionary accruals |DACC2|. For both measurements, we transform the residual values to the absolute values in which a higher value indicates a greater like-lihood that firms engage in earnings management hence, low accounting quality. Thus, the absolute value of discretionary accruals captures the combined effects of income-increasing and income-decreasing earnings management decisions (Warfield et al., 1995). For additional analyses, we also use the unadjusted values for discretionary accruals.

We measure the extent of NAS using two measures. First, we use a dummy variable HI_NAS that takes value one for the fourth (high) quartile based on non-audit fees paid to the incumbent auditors and zero otherwise. Second, we use L_NAS measured using the natural logarithm of non-audit fees paid to the incumbent auditors. We posit that a large proportion of NAS to total fees raises concerns of impairment of auditor independence (Ruddock et al., 2006).

For PCON firms, various measures of political connections have been used in prior studies, most commonly based on government control through patronage (Hillman, 2005; Lester et al., 2008; Leuz & Oberholzer-Gee, 2006; Mitchell & Joseph, 2010; Yeh et al., 2013). This study identifies political connection firms by individuals who have a seat in parliament, ministries, state, government officer (Faccio, 2006), political parties (Chan, 2012), government-linked firms and government leaders (Fung et al., 2015; Kamarudin et al., 2021; Wong & Hooy, 2020). This study extends prior studies by using a dummy variable DUM_POL that takes a value of one when at least one of the directors is a politician and zero otherwise (Fung et al., 2015; Hashmi et al., 2018). We used NUM_POL, measured by the number of politicians on the board of directors, for the alternative measurement.

3.3. Regression model

Our basic specification for testing the hypotheses is as follows:

(1) $\begin{array}{rl}& \begin{array}{c}\left|DACC\right|={\alpha }_0+{\beta }_{1}NASVAR+{\beta }_{2}POLVAR+{\beta }_{3}POLVAR\ast NASVAR\\ +{\beta }_{4}FSIZE+{\beta }_{5}ROA+{\beta }_{6}LEV+{\beta }_{7}GROWTH+{\beta }_{8}BIG4\\ +{\beta }_{9}CINT+{\beta }_{10}INVINT+{\beta }_{11}RDINT+{\beta }_{12}MKTBK\end{array}\\ & +{\delta }_{1-n}Industryeffects+{\theta }_{1-n}Yeareffects+{\epsilon }_{it}\end{array}$(1)

Where |DACC| is the absolute value of residual either |DACC1| or |DACC2|; |DACC1| is the absolute value of residual generated from the (Kothari et al., 2005) performance Adjusted model estimated based on Fama and French (1997) 10 industry groups; |DACC2| is the absolute value of residual generated from the modified-Jones performance adjusted model (Dechow et al., 1995) estimated based on Fama and French (1997) 10 industry groups; NASVAR is non-audit service variables either HI_NAS or L_NAS; HI_NAS is a dummy variable that takes value one for the fourth (high) quartile of based on non-audit fees paid to the incumbent auditors and zero otherwise; L_NAS is the natural logarithm of non-audit fees paid to the incumbent auditors; POLVAR is the political connection variable either DUM_POL or NUM_POL; DUM_POL is a dummy variable that takes value one if at least one of the directors is politician and zero otherwise; NUM_POL is the number of politicians on the board of directors; FSIZE is the natural logarithm of total assets; ROA is the ratio of net income to the total assets; LEV is the ratio of total debts divided by total assets; GROWTH is the change of annual net sales over last year sales; BIG4 is a dummy variable that takes a value of one if the firm is audited by one of the Big Four auditors and zero otherwise; CINT is the ratio of net book value of property, plant and equipment to total assets; INVINT is the ratio of total inventories to total sales; RDINT is the ratio of research and development expenditure to total assets; MKTBK is the ratio of market to book value; Fixed_Effects is a vector of industry and year.

4. Results

4.1. Descriptive statistics

Table 2 reports the descriptive statistics for all the variables. The results in Panel A of Table 2 indicated that the average |DACC1| is 0.386, with values ranging from 0.000 to 4.334. For |DACC2|, the average value is 0.397, with values ranging from 0.000 to 4.440. The dummy variable DUM_POL has a mean value of 0.453, meaning that 45.3% of the sample consisted of PCON firms. The NUM_POL has a mean value of 0.647, ranging from 0.000 to 7.000, indicating that the maximum number of politicians on board is seven directors. For NAS variables, the mean value for L_NAS is 2.280, while the dummy variable HI_NAS has a mean value of 0.239, indicating the higher quartile of NAS.

Table 2. Descriptive statistics

Panel A: Firm-level variables
Variable		Obs	Mean			Std.Dev.		Min		Max
|DACC1|		2245	0.386			0.452		0.000		4.334
|DACC2|		2245	0.397			0.464		0.000		4.440
DUM_POL		2245	0.453			0.498		0.000		1.000
NUM_POL		2245	0.647			0.879		0.000		7.000
HI_NAS		2245	0.239			0.427		0.000		1.000
L_NAS		2245	02.28			1.997		−1.204		8.191
FSIZE		2245	18.613			1.371		15.459		23.865
ROA		2245	0.034			0.106		−2.724		0.525
LEV		2245	0.178			0.154		0.000		0.877
GROWTH		2245	0.077			1.141		−1.000		26.160
BIG4		2245	0.471			0.499		0.000		1.000
CINT		2245	0.332			0.218		0.000		0.944
INVINT		2245	0.148			0.132		0.000		0.856
RDINT		2245	0.000			0.000		0.000		0.001
MKTBK		2245	1.482			3.955		0.013		80.84
Panel B: Univariate differences between variables of PCON and non-PCON firms
Variable	Non-PCON (N = 1228)				PCON (N = 1017)				p-values for differences
	Mean			SD	Mean		SD		diff		t
|DACC1|	0.424			0.424	0.340		0.480		0.087***		(4.510)
|DACC2|	0.436			0.435	0.350		0.492		0.090***		(4.530)
NUM_POL	0.000			0.000	1.429		0.767		−1.427***		(−59.529)
HI_NAS	0.216			0.412	0.267		0.443		−0.054**		(−2.978)
L_NAS	2.179			1.913	2.402		2.088		−0.243**		(−2.833)
FSIZE	18.385			1.273	18.888		1.434		−0.517***		(−8.908)
ROA	0.034			0.123	0.035		0.081		−0.001		(−0.251)
LEV	0.165			0.149	0.194		0.158		−0.029***		(−4.491)
GROWTH	0.057			0.823	0.100		1.434		−0.043		(−0.841)
BIG4	0.471			0.499	0.472		0.499		−0.004		(−0.171)
CINT	0.337			0.206	0.325		0.231		0.013		(1.347)
INVINT	0.157			0.134	0.136		0.130		0.022***		(3.955)
RDINT	0.000			0.000	0.000		0.000		−0.000**		(−3.208)
MKTBK	1.631			4.928	1.301		2.274		0.301		(1.900)

Notes: t statistics in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. See, Table 1 for the variable definitions.

For the control variables, the mean for FSIZE (the natural logarithm for total assets) is 18.613, with a range of 15.459 to 23.865. ROA has a mean value of 0034, with a minimum of −2.724 and a maximum of 0.525. For the variables LEV and GROWTH, the mean values are 0.178 and 0.077, respectively. The dummy variable BIG4 has a mean value of 0.471, indicating that Big 4 auditors audited 47.1 per cent of the sample. We also found that the mean values for CINT, INVINT, RDINT and MKTBK are 0.332, 0.148, 0.000, and 1.482.

Panel B of Table 2 reports the comparison of means for all the variables for the two groups, i.e., PCON and non-PCON firms. Sample of PCON firms exhibits lower NAS (HI_NAS and L_NAS), firm leverage (LEV), and R&D intensity (RDINT) than non-PCON firms. However, compared to non-PCON firms, the sample of PCON firms has higher discretionary accruals, |DACC1| and |DACC2|, and inventory intensity (INVINT).

We perform pair-wise correlation analysis among the dependent and independent variables. The untabulated results, for brevity purposes, show a high correlation among dependent variables where the correlation between |DACC1| and |DACC2| is 0.990. The correlation between DUM_POL and NUM_POL is 0.810, while the correlation between HI_NAS and L_NAS is 0.769. We find a strong correlation between |DACC1| and FSIZE is −0.729. Other variables, namely ROA, LEV, and BIG, have negative correlations with |DACC1| while INVINT has a positive correlation. Although the results indicate several significant correlations between independent variables, none represent any concern for multicollinearity.¹

4.2. Main empirical results

Table 3 presents the regression estimates of NAS and PCON on discretionary accruals for various subsamples. The coefficients for HI_NAS in columns (1) and (2) are positive and statistically significant, indicating that firms with high NAS exhibit a greater tendency to manage earnings than firms with low NAS. The result also shows that the coefficients for PCON in columns (1) and (2) are also positive and statistically significant, showing higher discretionary accruals in PCON firms than non-PCON firms. This result offers evidence supporting the notion that higher purchases of NAS could violate auditor independence and create agency problem (Ashbaugh et al., 2003) which lead to a reduced audit and accounting quality (Alareeni, 2019; Becker et al., 1998), but contradict the arguments supporting the knowledge spillover and audit efficiency effects (Parkash & Venable, 1993; Walker & Hay, 2013).

Table 3. Regression estimates of the effect of politically connected firms and the provision of non-audit services on discretionary accruals

	(1)	(2)
Variable	|DACC1|	|DACC1|
Intercept	5.061***	5.066***
	(42.746)	(42.827)
DUM_POL	0.035***	0.052***
	(2.629)	(3.445)
HI_NAS	0.050***	0.085***
	(2.933)	(3.763)
DUM_POL *HI_NAS		−0.072**
		(−2.360)
FSIZE	−0.252***	−0.252***
	(−41.713)	(−41.816)
ROA	−0.069	−0.064
	(−1.024)	(−0.949)
LEV	−0.044	−0.045
	(−0.940)	(−0.971)
GROWTH	0.010*	0.011*
	(1.841)	(1.875)
BIG4	−0.031**	−0.030**
	(−2.074)	(−1.994)
CINT	0.053	0.048
	(1.466)	(1.337)
INVINT	−0.099*	−0.104*
	(−1.789)	(−1.883)
RDINT	697.344***	709.997***
	(4.801)	(4.890)
MKTBK	0.004**	0.004**
	(2.363)	(2.275)
Industry effects	Yes	Yes
Year effects	Yes	Yes
Adj. R^2	0.55	0.55
N	2245	2245
F-stat	110.345	106.534

*, ** and *** represent significance at p < 0.10, <0.05 and <0.01, respectively. t-values are reported in the parentheses.

In column (2), we find the coefficient for DUM_POL*HI_NAS is negative and significant, suggesting that in the PCON sample, NAS has a lower positive impact on discretionary accruals than the non-PCON sample. Despite the negative perception of the consequences of PCON, this study supports the proposition that the existence of politicians on board adds additional monitoring mechanisms, especially in managing the adverse effects of NAS. Since firms must obtain approval from directors and audit committee before purchasing NAS (MCCG, 2017), having politicians on the board would significantly influence this decision and further reduce the Type I agency problem between principal and agent. Our findings add to the literature on the positive outcomes of PCON, inter alia, high share prices (C. M. Chen et al., 2013), market protection (Faccio, 2010), preferential access to credit (Y. Chen et al., 2020) and future performance (Azmi et al., 2020; Marzuki & Shukri, 2019). In sum, our findings support the proposition that the politician’s appointment aligns with the resource dependency theory (Chung et al., 2019) showed by the PCON firms’ ability to alleviate the negative effect of NAS.

In summary, the results in Table 3 suggest that high provision of NAS would lead to low accounting quality, exhibited by high discretionary accruals, consistent with prior studies (Al-Dhamari & Ku Ismail, 2015; Chaney et al., 2011; Hashmi et al., 2018). However, it is vital to note that such effects vary across PCON and non-PCON firms. In this study, we suggest that PCON firms would marginally alleviate the quality of accounting information compared to non-PCON firms. However, the substitutive effect between these two variables shows that the negative effect of NAS on accounting quality, exhibited by high discretionary accruals, is weaker in PCON firms as it is subject to higher discretionary accruals.

4.3. Self-selection and endogeneity

In our primary analysis, the endogeneity problem is related to selection bias because PCON firms could be systematically different from non-PCON firms. For example, PCON firms might self-select into the sample of highly |DACC1| or |DACC2|. To correct the problem, we utilised the propensity score matching (PSM) technique to control for firm-level characteristics, as developed by Rosenbaum and Rubin (1983).²

To evaluate treatment effects, we matched PCON firms to a set of control firms with non-PCON firms, with characteristics with the closest propensity score to firms with PCON firms.³ First, we estimate equation (2)(2) $Pr\left[PCO{N}_{it}=1\right]={\beta }_0+{\beta }_{1}FSIZ{E}_{it}+{\beta }_{2}LEV{i}_t+{\beta }_{3}BIG{4}_{it}+{\beta }_{4}CIN{T}_{it}+{\beta }_{5}INVIN{T}_{it}+{\beta }_{6}MKTB{K}_{it}+{\delta }_{1-n}Fixed\mathrm{\_}effects+{\epsilon }_{it}$(2) , a propensity score model that estimates the probability that a PCON firm is based on observable firm-specific variables. We then applied a condition on the highest propensity calliper to remove dissimilar matched pairs if the propensity scores (probabilities) difference is greater than 0.001.

In equation (2)(2) $Pr\left[PCO{N}_{it}=1\right]={\beta }_0+{\beta }_{1}FSIZ{E}_{it}+{\beta }_{2}LEV{i}_t+{\beta }_{3}BIG{4}_{it}+{\beta }_{4}CIN{T}_{it}+{\beta }_{5}INVIN{T}_{it}+{\beta }_{6}MKTB{K}_{it}+{\delta }_{1-n}Fixed\mathrm{\_}effects+{\epsilon }_{it}$(2) , we incorporated FSIZE, LEV, BIG4, CINT, INVINT, and MKTBK, consistent with prior studies (Abbott et al., 2003; Firth et al., 2009). For example, Abbott et al. (2003) find that NAS is higher for larger firms and clients of big audit firms. To mitigate the confounding effects, we also included the industry and year fixed effects, as presented below:

(2) $Pr\left[PCO{N}_{it}=1\right]={\beta }_0+{\beta }_{1}FSIZ{E}_{it}+{\beta }_{2}LEV{i}_t+{\beta }_{3}BIG{4}_{it}+{\beta }_{4}CIN{T}_{it}+{\beta }_{5}INVIN{T}_{it}+{\beta }_{6}MKTB{K}_{it}+{\delta }_{1-n}Fixed\mathrm{\_}effects+{\epsilon }_{it}$(2)

where all the variables are as defined earlier.

The results in column (1) of Table 4 present the logit regression estimates for equation (4), which estimate PCON firms’ probability. FSIZE and LEV are positively related to the probability of a PCON. We also find the probability of a PCON to have negative associations with BIG4, CINT, INVINT, and MKTBK.

Table 4. Regression estimates using the propensity score matching procedure

First Stage		Second Stage
	(1)		(2)	(3)	(4)
Variable	DUM_POL	Variable	|DACC1|	|DACC1|	|DACC1|
Intercept	−5.106***	Intercept	5.736***	5.693***	5.745***
	(−7.170)		(35.366)	(35.397)	(35.351)
FSIZE	0.289***	DUM_POL	0.063***	0.079***
	(7.408)		(3.375)	(3.204)
LEV	0.572*	HI_NAS	0.123***		0.095***
	(1.885)		(4.082)		(3.445)
BIG4	−0.334***	DUM_POL*HI_NAS	−0.138***
	(−3.421)		(−3.422)
CINT	−0.514**	L_NAS		0.024***
	(−2.431)			(3.611)
INVINT	−1.381***	DUM_POL*L_NAS		−0.022**
	(−3.878)			(−2.519)
MKTBK	−0.039**	NUM_POL			0.032***
	(−2.425)				(2.726)
		NUM_POL*HI_NAS			−0.056***
					(−2.595)
		ROA	0.049	0.044	0.050
			(0.622)	(0.563)	(0.637)
		GROWTH	0.020***	0.020***	0.020***
			(2.740)	(2.674)	(2.705)
		FSIZE	−0.288***	−0.286***	−0.288***
			(−34.647)	(−34.759)	(−34.553)
		LEV	−0.047	−0.042	−0.043
			(−0.769)	(−0.696)	(−0.704)
		BIG4	−0.022	−0.029	−0.023
			(−1.139)	(−1.475)	(−1.205)
		CINT	0.001	0.005	0.001
			(0.026)	(0.111)	(0.021)
		INVINT	−0.063	−0.054	−0.064
			(−0.896)	(−0.767)	(−0.907)
		MKTBK	0.006	0.005	0.005
			(1.239)	(1.085)	(1.023)
Industry effects	Yes	Industry effects	Yes	Yes	Yes
Year effects	Yes	Year effects	Yes	Yes	Yes
Pseudo R^2	0.0348	Adj. R^2	0.56	0.55	0.55
N	2230	N	1438	1438	1438
LR chi2	106.74	F-stat	72.734	72.336	72.142

* and *** represent significance at p < 0.10, <0.05 and <0.01, respectively. t-values are reported in the parentheses.

In columns (2) to (4), the regression estimates on the reduced sample, comprising 1,438 observations consisting of 719 firm-year observations from the PCON sample and 719 firm-year observations from the non-PCON sample, show that our inferences remain unchanged. We observe significant positive coefficients for DUM_POL in columns (2) and (3) and also for NUMPOL, an alternative measure for PCON, in column (4), exhibiting lower accounting quality in PCON firms compared to non-PCON firms. For proxies for NAS, in column (2), we find the coefficients for HI_NAS are significantly positive, similar to column (3), where the coefficient for L_NAS is also positive. The results support our earlier findings that NAS is associated with high discretionary accruals, i.e., low accounting quality. We observe that the coefficients for PCON*HI_NAS, PCON*L_NAS, and NUM_POL*HI_NAS, as reported in columns (2) to (4), respectively, are consistent with the main findings. From the analyses, we find robust evidence of differences in discretionary accruals between firms with high and low NAS and attest to the moderating effect of PCON.

4.4. Robustness tests

We perform a battery of tests to ensure our findings are robust to various specifications. First, we employ alternative measurements for discretionary accruals where we use the absolute value of residual generated from the modified-Jones performance-adjusted model (Dechow et al., 1995) estimated based on Fama and French (1997) industry groups. Second, we use alternative measures for NAS. Using the natural logarithm for NAS (L_NAS), we estimated equation (2)(2) $Pr\left[PCO{N}_{it}=1\right]={\beta }_0+{\beta }_{1}FSIZ{E}_{it}+{\beta }_{2}LEV{i}_t+{\beta }_{3}BIG{4}_{it}+{\beta }_{4}CIN{T}_{it}+{\beta }_{5}INVIN{T}_{it}+{\beta }_{6}MKTB{K}_{it}+{\delta }_{1-n}Fixed\mathrm{\_}effects+{\epsilon }_{it}$(2) . Third, we employ NUMPOL, the number of politicians on board, as a proxy for PCON. The untabulated results, for brevity purposes, show qualitatively similar findings. We find the coefficient NAS, L_NAS, HI_NAS are positive, implying reduced accounting quality -high discretionary accruals and the coefficients for the interaction variables DUM_POL*HI_NAS, DUM_POL*L_NAS, HI_NAS*NUM_POL, NUMPOL*HI_NAS are negative, consistent with the notion that the relationship is weakened in PCON firms.

Finally, we further test whether the relationship between NAS and |DACC| is non-linear with a U-shaped curve. We include LNAS and LNAS² in the equation estimated on the PCON, non-PCON and pooled samples. If this prediction is true, the coefficient for LNAS is expected to be positive (negative), while the coefficient for LNAS² is negative (positive). The results show that the coefficient for LNAS², the quadratic form, is negative and significant, while the coefficient for LNAS is positively significant. These results imply a non-linear association between NAS provision and discretionary accruals. Specifically, we find that the provision of NAS results in high discretionary accruals or lower accounting quality and reverses after the optimal point. These results are consistent with the knowledge spillover effect occurring at high levels of NAS. In contrast, the results suggest that the potential for issues related to economic bonding occurs at low levels of NAS until the optimal point. In general, our results exhibit the non-linear effect of NAS on accounting quality in the PCON, non-PCON and pooled sample, corroborating evidence by Beardsley et al. (2021), who test the non-linear effect of NAS on audit quality.

5. Conclusions

This study examines the effects of PCON and NAS on discretionary accruals. A battery of tests indicates high discretionary accruals, or low accounting quality, in firms with high NAS and PCON. However, the negative effect of NAS on discretionary accruals is more apparent in non-PCON firms. Though this study finds high discretionary accruals in PCON firms, the negative effect of NAS on accounting quality is weak in these firms. This finding offers a new explanation that PCON firms have better monitoring of NAS to maintain an acceptable level of discretionary accruals. Though PCON firms exhibit lower accounting quality corroborating the findings of (Mohamad et al., 2012), the PCON directors have a careful mindset not to permit the severe effect of NAS on discretionary accruals. This evidence probably supports the effectiveness of the initiative taken by the regulatory bodies to enhance the quality of financial reporting, such as the government transformation agenda.

Our results should be interpreted with caution despite the use of various analyses. Our sample was limited to the available data and written disclosure in the corporate reports. The influence of politicians might extend beyond board members through proxies or informal networking. Despite this limitation, our study provides valuable insights for investors and policymakers to understand the effect of NAS and PCON on accounting quality. To make further progress, we encourage more research on the impact of different types of NAS and PCON. In summary, this study a new perspective on the uniqueness of business-political nexuses.

Acknowledgement

The authors thank the workshop participants at the Universitas Riau, Indonesia and Universiti Teknologi Mara, Malaysia for insightful and constructive comments. Arumega Zarefar would like to thank Indonesia Endowment Fund for Education (LPDP) from the Ministry of Finance Republic Indonesia for granting the scholarship.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The authors received no direct funding for this research.

Notes

1. Multicollinearity is likely to be a concern when pair-wise correlation between the two variables exceeds 0.80 (Gujarati, 1995).

2. Shipman et al. (2017) argued that propensity score matching does not address most concerns relating to self-selection or endogeneity, hence it is inaccurate to suggest that the procedure is an alternative to Heckman (1979) type selection models.

3. Compared to other methods of controlling for endogeneity, the benefits of PSM is that it does not rely on a clear source of identification of exogenous variables (Roberts & Whited, 2013).

Appendix 1:

Variable description

Table

Variable	Definition
Dependent Variables
|DACC1|	The absolute value of residual generated from the Kothari et al. (2005) performance adjusted model estimated based on Fama and French’s (1997) 10 industry groups
|DACC2|	The absolute value of residual generated from the modified-Jones performance adjusted model (Dechow et al., 1995) estimated based on Fama and French’s (1997) 10 industry groups
Independent Variable
HI_NAS	A dummy variable that takes value one for the fourth (high) quartile of based on non-audit fees paid to the incumbent auditors and zero otherwise.
L_NAS	The natural logarithm of non-audit fees paid to the incumbent auditors.
Moderating variable
DUM_POL	A dummy variable that takes value one if at least one of the directors is politician and zero otherwise.
NUM_POL	The number of politicians on the board of directors.
Control variables
FSIZE	The natural logarithm of total assets
ROA	The ratio of net income to the total assets
LEV	The ratio of total debts divided by total assets
GROWTH	The change of annual net sales over last year sales
BIG4	A dummy variable that takes a value of one if the firm was audited by one of the Big Four auditors and zero otherwise.
CINT	The ratio of net book value of property, plant and equipment to total assets
INVINT	The ratio of total inventories to total sales.
RDINT	The ratio of research and development expenditure to total assets
MKTBK	The ratio of market to book value

Appendix 2:

Pairwise Correlation Matrix

Table

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
(1) |DACC1|	1.000
(2) |DACC2|	0.990*	1.000
(3) DUM_POL	−0.092*	−0.093*	1.000
(4) NUM_POL	−0.138*	−0.139*	0.810*	1.000
(5) HI_NAS	−0.237*	−0.237*	0.060*	0.116*	1.000
(6) L_NAS	−0.234*	−0.234*	0.056*	0.103*	0.769*	1.000
(7) FSIZE	−0.729*	−0.729*	0.183*	0.259*	0.367*	0.344*	1.000
(8) ROA	−0.110*	−0.110*	0.004	0.003	0.077*	0.073*	0.128*	1.000
(9) LEV	−0.237*	−0.238*	0.092*	0.095*	0.102*	0.081*	0.318*	−0.179*
(10) GROWTH	−0.005	−0.004	0.019	0.014	0.014	0.031	0.046	0.077*
(11) BIG4	−0.319*	−0.320*	0.001	0.079*	0.372*	0.433*	0.406*	0.063*
(12) CINT	−0.015	−0.012	−0.028	−0.051	0.057*	0.029	0.021	−0.085*
(13) INVINT	0.068*	0.067*	−0.082*	−0.110*	−0.070*	−0.059*	−0.145*	0.032
(14) RDINT	−0.054	−0.055*	0.074*	0.080*	0.032	0.015	0.173*	−0.002
(15) MKTBK	−0.049	−0.050	−0.041	−0.034	0.116*	0.120*	0.115*	0.285*
Variables	(9)	(10)	(11)	(12)	(13)	(14)	(15)
(9) LEV	1.000
(10) GROWTH	0.005	1.000
(11) BIG4	0.077*	−0.040	1.000
(12) CINT	0.025	−0.044	0.121*	1.000
(13) INVINT	0.099*	0.007	−0.133*	−0.295*	1.000
(14) RDINT	0.092*	−0.009	0.035	0.052	−0.042	1.000
(15) MKTBK	−0.008	0.007	0.096*	0.033	−0.068*	0.014	1.000

*shows significance at the .01 level

Appendix 3: Robustness Analysis: Regression estimates of the effect of politically connected firms and the provision of non-audit services on discretionary accruals

Table

	(1)	(2)	(3)	(4)	(5)
Variable	|DACC2|	|DACC1|	|DACC1|	|DACC1|	|DACC2|
Intercept	5.195***	5.020***	5.148***	5.108***	5.238***
	(42.842)	(42.793)	(42.804)	(42.741)	(42.752)
DUM_POL	0.053***	0.061***	0.062***
	(3.422)	(3.038)	(3.024)
HI_NAS	0.088***			0.078***	0.080***
	(3.792)			(3.721)	(3.749)
DUM_POL*HI_NAS	−0.074**
	(−2.373)
L_NAS		0.013***	0.013***
		(2.611)	(2.636)
DUM_POL *L_NAS		−0.011*	−0.012*
		(−1.714)	(−1.726)
NUM_POL				0.039***	0.040***
				(4.113)	(4.089)
NUM_POL*HI_NAS				−0.038**	−0.039**
				(−2.425)	(−2.435)
FSIZE	−0.259***	−0.250***	−0.256***	−0.254***	−0.261***
	(−41.834)	(−41.786)	(−41.802)	(−41.712)	(−41.727)
ROA	−0.068	−0.066	−0.069	−0.064	−0.068
	(−0.977)	(−0.970)	(−0.998)	(−0.952)	(−0.980)
LEV	−0.047	−0.043	−0.044	−0.044	−0.046
	(−0.974)	(−0.916)	(−0.919)	(−0.949)	(−0.953)
GROWTH	0.011**	0.011*	0.011*	0.011*	0.011**
	(1.974)	(1.860)	(1.958)	(1.863)	(1.961)
BIG4	−0.031**	−0.031**	−0.032**	−0.032**	−0.033**
	(−2.042)	(−1.995)	(−2.044)	(−2.123)	(−2.170)
CINT	0.054	0.053	0.059	0.051	0.057
	(1.465)	(1.460)	(1.588)	(1.405)	(1.533)
INVINT	−0.110*	−0.097*	−0.103*	−0.097*	−0.103*
	(−1.934)	(−1.757)	(−1.806)	(−1.752)	(−1.803)
RDINT	725.257***	701.520***	716.559***	687.599***	702.150***
	(4.873)	(4.824)	(4.807)	(4.742)	(4.724)
MKTBK	0.004**	0.004**	0.004**	0.004**	0.004**
	(2.225)	(2.288)	(2.238)	(2.351)	(2.300)
Industry effects	Yes	Yes	Yes	Yes	Yes
Year effects	Yes	Yes	Yes	Yes	Yes
Adj. R^2	0.55	0.55	0.55	0.55	0.55
N	2245	2245	2245	2245	2245
F-stat	106.732	105.900	106.091	106.915	107.106

*, ** and *** represent significance at p < 0.10, <0.05 and <0.01, respectively. t-values are reported in the parentheses.

Appendix 4: Regression estimates of the non-linear effect of the provision of non-audit services on discretionary accruals

Table

	(1)	(2)	(3)
Variable	PCON	NON-PCON	POOLED
Intercept	5.296***	5.748***	5.351***
	(26.498)	(36.200)	(43.441)
LNAS	−0.056***	−0.070***	−0.055***
	(−3.545)	(−5.658)	(−5.616)
LNAS^2	0.010***	0.019***	0.012***
	(3.602)	(7.886)	(6.953)
FSIZE	−0.261***	−0.287***	−0.266***
	(−25.480)	(−35.634)	(−42.511)
ROA	−0.328**	0.025	−0.060
	(−2.107)	(0.354)	(−0.866)
LEV	−0.082	−0.116**	−0.049
	(−1.008)	(−2.011)	(−1.036)
GROWTH	0.015*	0.011	0.012**
	(1.919)	(1.138)	(1.997)
BIG4	0.021	−0.060***	−0.029*
	(0.754)	(−3.321)	(−1.842)
CINT	−0.009	0.107**	0.064*
	(−0.156)	(2.308)	(1.736)
INVINT	0.042	−0.172***	−0.104*
	(0.443)	(−2.590)	(−1.852)
RDINT	618.746***	0.000	647.061***
	(3.646)	(0.000)	(4.370)
MKTBK	0.011**	0.002	0.003
	(2.005)	(1.166)	(1.645)
Industry effects	Yes	Yes	Yes
Year effects	Yes	Yes	Yes
Adj.R2	0.51	0.62	0.56
N	1017	1228	2245
F-stat	44.814	84.453	113.830

*, ** and *** represent significance at p < 0.10, <0.05 and <0.01, respectively. t-values are reported in the parentheses.