the smaller, the higher the degree of downward real 
surplus management. 
3.2.4  Controlled Variable 
Based on previous research results related to equity 
pledge  and  earnings  management,  the  following 
variables are used as control variables: (1) Nature of 
property rights (SOE): Based on whether it is a state-
owned enterprise, the value is 1 if it is a state-owned 
enterprise, and 0 if it is not. (2) Shareholding ratio of 
the largest shareholder (Top1): This variable is used 
to  represent  the  concentration  of  ownership.  It  is 
measured by the proportion of the shares held by the 
largest  shareholder  to  the  total  shares  of  the  listed 
company.  Generally,  the  larger  the  proportion,  the 
higher the degree of equity concentration. (3) Audit 
quality  (Big4):  The  Big  Four  accounting  firms  are 
relatively  authoritative  accounting  firms.  It  is 
measured according to whether the listed company is 
audited by the Big Four accounting firm at the end of 
the year. (4) Size of the company (Size): The size of 
the  company  is  measured  by  taking  the  natural 
logarithm of the company's total assets. The larger the 
indicator  is,  the  larger  the  company  is.  (5)  Growth 
rate  of  operating  income  (Growth):  It  is  used  to 
express the development ability of the enterprise. It is 
measured  by  dividing  the  operating  income  of  the 
current year by the operating income of the previous 
year  and  subtracting  one.  (6)  Number  of  directors 
(Board): Indicates the size of the board of directors of 
the listed company, which is measured by taking the 
natural  logarithm  of  the  number  of  directors  of  the 
company. (7) Tobin's Q value (Tobin): This variable 
is used to measure the company's performance. It is 
measured by the ratio of the company's market value 
to the replacement cost of assets. At the same time, 
two dummy variables of year and industry are defined 
to control their effects. 
3.3  Model Specification 
In this paper, the stepwise test method of mediation 
test is used to test whether the total effect coefficient 
is  significant,  that  is,  whether  there  is  a  significant 
relationship between the independent variable and the 
dependent  variable.  If  significant,  proceed  to 
subsequent  analyses,  if  not,  mediation  analysis  is 
terminated.  Test  whether  the  effect  of  the 
independent variable acting on the mediating variable 
is  significant;  if  it  is  significant,  continue  the 
subsequent test, otherwise stop the analysis, and the 
mediating effect does not exist; test whether the effect 
of  the  mediating  variable  acting  on  the  dependent 
variable  is  significant;  if  significant,  continue  the 
subsequent  test,  otherwise  stop  the  analysis,  the 
mediating effect does not exist; test whether the direct 
effect  is  significant.  When  the  first  two  items  are 
significant, if it is not significant, there is a complete 
mediation  effect  (Judd,  Kenny,  1981),  otherwise 
there  is  a  partial  mediation  effect  (Baron  &  Kenny 
1986). 
The following models are set up to conduct the 
mediation  effect  test  of  assumptions 1,2 and  3, and 
then use the Bootstrap method if the step-wise test is 
not significant. 
Hypothesis 1a、1b was verified to construct the 
following multiple linear regression model: 
Radinti,  t=β0+β1Pledgei,  t+β2Soei,  t+β3Roai, 
t+β4Top1i,  t+β5Big4i,  t+β6Sizei,  t+β7Growthi, 
t+β8Boaedi, t+β9Tobini, t+∑Y+∑IND+ε              (2) 
Hypothesis  2  was  verified  to  construct  the 
following multiple linear regression model: 
Pledgei,  t=β0+β1DREMi,  t+β2Soei,  t+β3Roai, 
t+β4Top1i,  t+β5Big4i,  t+β6Sizei,  t+β7Growthi, 
t+β8Boaedi, t+β9Tobini, t+∑Y+∑IND+ε              (3) 
To  verify  hypothesis  3,  the  following  multiple 
linear  regression  model  was  constructed  by 
combining the models (2) and (3): 
Radinti, t=β0+β1Pledgei, t+β2DREMi, t+β3Soei, 
t+β4Roai,  t+β5Top1i,  t+β6Big4i,  t+β7Sizei, 
t+β8Growthi,  t+β9Boaedi,  t+β10Tobini, 
t+∑Y+∑IND+ε     
(4) 
4  EMPIRICAL ANALYSIS 
4.1  Descriptive Statistics 
The descriptive statistics of the variables in this paper 
are shown in Table 2. The minimum value of R & D 
investment  of  listed  companies  is  0,  and  the 
maximum value is 251.6%. It can be seen that there 
is a large gap in the innovation level of different listed 
companies, and there is an extreme phenomenon of 
excessive  innovation  investment  in  that  year.  For 
example,  in  Junshi  Biology  in  2019  and  2020,  this 
value reached 122.06% and 112.72%. The median is 
3.4%, indicating that the overall innovation level of 
listed companies is low, and the innovation impetus 
is  insufficient.  The  average  equity  pledge  ratio 
(Pledge)  of  controlling  shareholders  is  44.5%, 
indicating  that  the  equity  pledge  situation  of 
controlling  shareholders  of  listed  companies  is 
relatively common. 
The  Variance  Inflation  Factor  test  refers  to  the 
ratio of the  variance when there is  multicollinearity 
BDEDM 2022 - The International Conference on Big Data Economy and Digital Management