Conceptual framework and related literature

Innovation at the firm level is driven primarily by research and development (R&D) and by various conditions that govern the business prospects. We argue that if a firm anticipates a highly corrupt bureaucracy, red tapes, and rent-seeking behavior on the part of the public authorities that independently or collectively hurts productivity and profitability, then these enterprises might choose not to innovate and grow. Coupled with this, firms may also face the problem of credit constraints that deter investments for R&D and therefore, independently as much as because of the impact of corruption, underperform on innovative abilities. The flow of information regarding perception of corruption and probable hindrances lead to adverse outcomes as much as these factors hurt innovation in their material forms. The issue at hand is information theoretic and includes effects of negative externality on firm behavior. Indeed, many studies have shown strong association between innovation activities and survival chances of firms (Cefis and Marsili, Reference Cefis and Marsili2005; Pérez et al., Reference Pérez, Llopis and Llopis2004; Ericson and Pakes, Reference Ericson and Pakes1995; Hall, Reference Hall1987, etc.). However, the conclusions with respect to innovation's effect on firm growth are ambiguous. For example, Hölzl (Reference Hölzl2009) does find that R&D efforts are bigger for high-growth firms, but the relationship is still dependent on the average technological capability of the country. Conversely, studies have also explored the effects of innovation on firm productivity (Mairesse and Mohnen, Reference Mairesse and Mohnen2010; Griffith et al., Reference Griffith, Huergo, Mairesse and Peters2006; Mohnen et al., Reference Mohnen, Mairesse and Dagenais2006; Griffith et al., Reference Griffith, Redding and Van Reenen2004).

When it comes to exploring the link between corruption and innovation, as suggested by Xu and Yano (Reference Xu and Yano2017), very few studies have explored the association. Innovators can be vulnerable to the presence of corruption especially involving government services because they depend on such services for permits and licenses (Murphy et al., Reference Murphy, Shleifer and Vishny1993). Some other studies, employing both micro and macro samples, that have tested the detrimental impact of corruption for innovative firms are Paunov (Reference Paunov2016), Waldemar (Reference de Waldemar2012), and Anokhin and Schulze (Reference Anokhin and Schulze2009).The outcome is not always negative, as Xu and Yano (Reference Xu and Yano2017) find that anticorruption efforts make Chinese firms invest more in R&D activities.

Presently, we explain why exploring this question offers significant importance for India. First of all, starting from the seminal paper of Krueger (Reference Krueger1974), corruption has been undeniably considered a major obstacle for growth and development outcomes of India. In more recent years, Heston and Kumar (Reference Heston and Kumar2008) have stressed the crucial role of corruption in affecting economic growth of India. But more topically, a few recent studies have explored corruption's impact on firm performance for India. For example, Sharma and Mitra (Reference Sharma and Mitra2015) use bribe data as a measure of corruption to show that while corruption affects firms’ efficiency negatively, it does help with respect to export and product innovation. However, with respect to the probability of innovation by firms facing high degrees of corruption across states in India, there is a serious gap in the literature. As mentioned in the preceding text, we explore this relation but also extend it beyond perceived corruption only. We consider other obstacles as viewed by firms starting with financial barriers and account for interactive effects of corruption with the other perceived obstacles. The idea is to see if other perceived obstacles accentuate the damaging impact of corruption perceptions on probability to innovate for firms. The infrastructure-related barriers like unsteady and fluctuating power supply, transport, and customs-related obstacles feature in this group.

Data and the sources

Dependent and independent variables

Our dependent variable is a dummy based on the question “In the last three years, has this establishmentFootnote ¹ introduced new products or services?” (WBES, 2014). The dummy takes the value 1 if the answer is yes; 0 otherwise. The summary statistics of our main variable of interest are presented in table 1. Out of the 9,281 observations, approximately 44 percent of data points take the value 1. Looking across industries, we find that some of the most innovative industries are leather (56 percent), hotels and restaurants (54 percent), garments (52 percent), electronics (57 percent), and so on.

Table 1: Summary Statistics

Note: Innovation is a dummy taking the value 1 “if in the last three years, the establishment has introduced new products or services?” Corr. (perceived) is a variable representing corruption perceptions of firms. The question asked is How much of an obstacle is corruption? Finance obs. (perceived)is a variable based on the question How much of an obstacle is access to finance? Firm (small) represents the dummy for small-sized firms, Firm (medium) indicates medium-sized firms and firm (large) indicates dummy for large firms. Cap city denotes if the firm is in the official capital city. Business city denotes if the firm is in the main business city or not. Export (sales %) denotes the percent of sales constituting of exports. Years in operation is the number of years the firm has been in operation. Part of large est. denotes if the firm is part of a larger company (establishment) or not.

States like Karnataka, Kerala, Rajasthan, and a few others have firms that are most innovative.

Regarding the main independent variable of interest, which is a measure of firms’ perception of corruption, we find that the specific question asked to firm owners is: How much of an obstacle is corruption? The question aims to assess the extent to which firm owners consider corruption as an obstacle to current operations of the establishment. The survey categorizes corruption as an obstacle in five levels. These are no obstacle, minor obstacle, moderate obstacle, major obstacle, and very severe obstacle. We construct an ordered dummy variable ranging from 1 to 4 with 0 indicating no obstacle and higher numbers representing greater perceived obstacle. The mean for our sample is around 2.1. About 33 percent of the firms categorize corruption as an obstacle to be 0 or 1 implying that they do not perceive corruption to be much of a hindrance. About 43 percent of firms in the sample state that they perceive corruption as a major or very severe obstacle in their operations. The average score for the variable in states like Uttar Pradesh, Uttaranchal, Punjab, Tamil Nadu, Rajasthan, and Odisha is relatively high (close to 3 or higher).

Other independent variables and controls

Next, access to finance stands out as an important perceived obstaclefor firms. We discuss shortly if access to finance and corruption are related. However, the variable measuring an obstacle to financial access is as similarly constructed as corruption. As stated in the survey, difficulty in accessing finance by firms includes availability as well as cost, interest rates, fees, and collateral requirements. An ordered dummy variable is constructed like before; it varies from 0 to 4, where 0 indicates responses stating no obstacle. Higher numbers indicate greater obstacle. The mean is smaller than that in the case of corruption (1.2) suggesting that for our sample firms in general perceived corruption as more of an obstacle relative to that of acquiring finance. Almost 85 percent of the firms report no obstacle, minor obstacle, or moderate obstacle.

Our benchmark control variables include firm size, location of the firm, percent of the firm's sales that are exports, if the firm belongs to a larger company, and number of years the firm is in business. As pointed out by Schumpeter (Reference Schumpeter1942), larger firms are more likely to engage in innovation activity because their chances of facing liquidity constraints are likely to be lower than small and medium-sized firms. At the same time, they can also enjoy economies of scale (Mairesse and Mohnen, Reference Mairesse and Mohnen2002; Cohen and Klepper, Reference Cohen and Klepper1996). We control for dummies indicating smallFootnote ² and medium firms with “large firm” dummy as the baseline. About 34 percent of the firms in our sample are small sized and about 44 percent are medium sized.

Another control incorporated here is the percentage of sales amounting to exports. As suggested by Pellegrino and Savona (Reference Pellegrino and Savona2017) and Narula and Zanfel (Reference Narula, Zanfei, Fagerberg, Mowery and Nelson2003), firms operating in the international market are more likely to innovate because it involves greater market competitiveness. The mean of our sample is about 6 percent with a high standard deviation suggesting that there is greater variation in terms of exports among the firms. Next, the conclusion in the literature with respect to impact of firm's age on innovation abilities is ambiguous. For example, Klepper (Reference Klepper1996) argued that innovations per firm should be higher for the younger firms, theoretically speaking. However, studies like Galende and De la Fuente (Reference Galende and De la Fuente2003) explain that a firm's knowledge and experience accumulated over time can play a positive role in innovation abilities. Recent studies like Pellegrino and Savona (Reference Pellegrino and Savona2017) and Bertoni and Tykvovà (Reference Bertoni and Tykvová2015) have considered a nonlinear impact of age. Based on survey information, we construct age of firms as being the number of years it is has been in operation as of 2014. The average age is about twenty years. We further control for dummy variables suggesting if the firm is located in the capital city and whether it is the main business city or not. Location factors bring advantages in the form of positive externalities through networking, proximity to lenders, and availability of skilled workers. All these, in turn, should increase the likelihood of innovation by firms. Apart from these, we control for state and industry fixed effects.

Empirical methodology

Benchmark specifications

The first part of our hypothesis explores how perception of corruption among firm owners in India affects their probability to innovate. The following logit specification is empirically tested:

(1)$${\rm Inn}{\rm o}_{{\rm ijs}} = {\rm \alpha }_0 + \alpha _1{\rm Cor}{\rm r}_{{\rm ijs}} + \delta Controls_{ijs}{\rm \rho }_{\rm i} + {\rm \theta }_{\rm s} + {\rm \epsilon }_{{\rm it}}$$

Where, Inno_ijs is the dummy variable suggesting if firm i in industry j in state s innovates or not. Corr _ijs is the perception of corruption by firm owners ranging from 0 to 4. The benchmark controls, as stated earlier, are firm size, firm location, export share in sales of firms, and years in operation. ρ_i represents the industry fixed effects and θ_s represent the state fixed effects. A negative coefficient of α₁ will imply that when firms perceive corruption as a stronger obstacle, their likelihood of innovation goes down. A positive coefficient will suggest otherwise.

The second part of the analysis aims at estimating the following equation

(2)$${\rm Inn}{\rm o}_{{\rm ijs}} = {\rm \alpha }_0 + \alpha _1{\rm Cor}{\rm r}_{{\rm ijs}} + \alpha _2{\rm Fi}{\rm n}_{{\rm ijs}} + \alpha _3( {Corr \, \ast \, Fin} ) _{ijs} + \delta Controls_{ijs}{\rm \rho }_{\rm i} + {\rm \theta }_{\rm s} + {\rm \epsilon }_{{\rm it}}\;$$

For the second part of our analysis, we are particularly interested in the coefficients: α₁, α₂, and α₃. Specifically, we are interested in estimating the overall impact of perceived corruption on probability to innovate. In the presence of interaction terms, the marginal estimate is given by ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}}} = {\rm \alpha }_1 + {\rm \alpha }_3Fin_{{\rm ijs}}$. Whether ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}} }> 0$ or <0 will depend on the sign and magnitude of α₁ and α₃ as well as on the magnitude of Fin _ijs. If for greater values of Fin _ijs, ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}} }< 0$ that would imply that as firms perceive greater obstacles in accessing finance, a rise in corruption perception lowers the probability to innovate. For greater values of Fin _ijs, ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}} }> 0$ will imply the opposite. As described in the “Robustness Analysis,” we also interact perceived corruption with other nonfinancial obstacles as perceived by firms.

Empirical methodology

Our benchmark model consists of logit specifications. Studies dealing with binary dependent variables and considering firm-level analysis have usually employed limited dependent variable (LDV) models like probit or logit (Webster and Piesse, Reference Webster and Piesse2018; Swamy et al., Reference Swamy, Knack, Lee and Azfa2001). Ordinary least square (OLS), under these circumstances, suffer from challenges like predicted probabilities lying outside the unit interval. Logit estimators, similar to probit, use Maximum Likelihood Estimation (MLE). But they use a logistic distribution function of the error terms.

The initial specification can be written as

(3)$$\Pr ( {Inno = 1} \hskip-.3pt) = F\;( \hat{X}{\rm \Omega }) $$

$\Pr ( {Inno{\rm \;} = 1} )$ is the probability whether a firm innovates or not. While F is the cumulative standard logistic distribution, X is the vector of explanatory variables and Ω is the vector of coefficients to be estimated. When firms do respond yes to innovate, the event is categorized as a success. y* is the latent continuous metric that underlies theobserved responses by the analyst. Firm's probability to innovate will depend on an unobservable latent (utility) index I _i which, in turn, is determined by an array of explanatory variables. Finally, the model we estimate can be written as $\Pr ( {Inno_{ijs} = 1{\rm \vert }X_{ijs}} ) = {\rm \;\Phi }( {\beta X_{ijs}} )$. We formulate our hypothesis in the equations (1) and (2) that is empirically tested using a logit fixed effect model. Fixed effects take into account invariant industry and state fixed effects. As we know that for logit regressions, the estimated coefficients cannot be meaningfully interpreted. We, thus, report marginal estimates for all the variables because average coefficients have the potential to be biased (Webster and Piesse, Reference Webster and Piesse2018; Fernández-Val, Reference Fernandez-Val2009). Additionally, as elaborated later we also need to estimate marginal effects for concluding meaningfully the overall impact of corruption on probability to innovate (Dutta and Sobel, Reference Dutta and Sobel2021; Berry et al., Reference Berry, Golder and Milton2012; Braumoeller Reference Braumoeller2004). After presenting our benchmark results, we elaborate on our identification strategies in “Robustness Analysis.”

Benchmark results

In table 2, we present our first set of benchmark results corresponding to equation (1). We present the logit estimates along with the marginal effects in table 2. In column (1), we start with exploring a binary relationship by including only the variable of interest: perceived corruption of firms. We control for industry and state fixed effect in column (1). Based on marginal estimates in column (2), a unit rise in corruption score reduces probability to innovate by firms by 1 percent.

Table 2: Logit Specifications: Probability to Innovate for Firms and Perceived Corruption

Note: The dependent variable is Innovation: a dummy taking the value 1 “if in the last three years, the establishment has introduced new products or services?” Corr. (perceived) is our variable of interest representing corruption perceptions of firms. It is coded from 0 to 4. The question asked is How much of an obstacle is corruption? Firm (small) represents the dummy for small-sized firms and Firm (medium) indicates medium-sized firms. Capital city denotes if the firm is in the official capital city. Business city denotes if the firm is in the main business city or not. Export (sales %) denotes the percent of sales constituting of exports. Years in operation is the number of years the firm has been in operation. Part of large co. denotes if the firm is part of a larger company (establishment) or not. Robust Standard Errors are in parenthesis. We control for state and industry fixed effects.

***p < 0.01; **p < 0.05; *p < 0.1.

In column (3), we add all the benchmark controls. The marginal estimates corresponding to the specification in column (3) are presented in column (4). The probability fora unit rise in corruption perception score remains the same. But the significance of the coefficient of corruption perception becomes stronger after the inclusion of benchmark controls. In terms of controls, smaller firms are 16 percent less likely to innovate relative to large firms. The probability is about 4 percent for medium-sized firms. Being part of larger firm enhances the probability to innovate by about 5 percent. We also find that location of the firm significantly helps in terms of its probability to innovate. Thus, in general, we find that the sign and significance of the control variables add support to existing findings.

In table 3, we estimate specification (2). In column (1), we rerun specification from table 1 and include another perceived obstacle for firm owners—that of acquiring finance. We find that the variable measuring perceived difficulty in accessing finance is not significant while corruption perception continues to remain significant. This is also reflected in the marginal estimates in column (2). In column (3), we introduce the interaction term, Corr*Fin over the previous specification. As evident from the results, corruption perception loses its significance and the interaction term is negative and significant. Thus, it would imply that in presence of greater perceived obstacle about acquiring finances, a firm is less likely to innovate when its corruption perception rises. Yet, in the presence of interaction terms, the individual significance of the coefficients is not the appropriate consideration. The joint significance of the terms matter in this case. As Brambor, Clark, and Golder (Reference Brambor, Clark and Golder2006) point out, the magnitude and statistical significance usually vary for different values of the interacted variables (Dutta and Sobel, Reference Dutta and Sobel2021). For example, it is quite possible that the combination of effects is statistically significant or insignificant at different values of perceived difficulty in accessing finance, regardless of the individual significance of either coefficient in the regression. Thus, we need to examine the conditional effect for a range of values of the perceived obstacle in accessing finance (Dutta and Sobel, Reference Dutta and Sobel2021; Berry et al., Reference Berry, Golder and Milton2012; Braumoeller, Reference Braumoeller2004). Additionally, because our specifications are logit estimates, we need to calculate the marginal effects for all the variables.

Table 3: Logit Specifications: Probability to Innovate for Firms and Perceived Corruption, and Accessing Finance Obstacles (Perceived)

Notes: The dependent variable is Innovation: a dummy taking the value 1 “if in the last three years, the establishment has introduced new products or services?” Corruption is our variable of interest representing corruption perceptions of firms. It is coded from 0 to 4. The question asked is How much of an obstacle is corruption? Access to fin. (perceived) is our second variable of interest. It assesses perceived difficulties by firm owners in accessing finance based on the question How much of an obstacle is access to finance? It is coded from 0 to 4 with higher numbers representing stronger perceptions. Fin. Dummy is constructed based on access to finance, taking 1 for above averages values for the sample, 0 otherwise. Firm (small) represents the dummy for small-sized firms and Firm (medium) indicates medium-sized firms. Capital city denotes if the firm is in the official capital city. Business city denotes if the firm is in the main business city or not. Export (sales %) denotes the percent of sales constituting of exports. Years in operation the number of years the firm has been in operation. Part of large co. denotes if the firm is part of a larger company (establishment) or not. Robust Standard Errors are in parenthesis. We control for state and industry fixed effects.

(#) the marginal effects with respect to corruption are given at the end of the table.

***p < 0.01; **p < 0.05; *p < 0.1.

The marginal estimates for column (3) specification are presented in column (4). ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}}}$ is estimated for the minimum and maximum values of access to finance (perceived) score and the results are presented at the bottom of the table. We find that when firms perceive no obstacle in accessing finance, a rise in perceived corruption does not affect their probability to innovate. But for very severe perceived difficulties in accessing finance, a rise in perceived corruption lowers their probability to innovate by 4 percent. Thus, our analysis so far strongly suggests that perceived corruption is what firms care for when it comes to their decision to innovate or not. Higher perceived corruption lowers the probability to innovate. But perceived difficulty in accessing finance is not a hindrance on its own. Stronger perception about difficulties in acquiring finance affects a firm's probability to innovate through the channel of perceived corruption. Most of the controls retain the expected signs and significance in terms of the marginal effects as evident in column (4).

Note that, perceptions of firms about obstacles can potentially be correlated. For example, perception about corruption can be correlated with a firm's experience in terms of acquiring finance. We address this in three possible ways. First, while it is true that many borrowers have to pay bribes to access credit from financial institutions in India, especially from those owned and managed by public authorities, a large private sector has emerged since the 1990s that extend credit lines often at higher rates of interest and hidden charges, but mostly free of problems with rent seeking. We have shown that corruption perception by itself (negative significant, Table 2) hurts innovation. Subsequently, when we include access to finance (Table 3, positive significant) it helps innovation, but renders corruption perception insignificant. However, its interaction with corruption (Table 3, negative significant) is again damaging for innovation, while corruption perception is itself statistically insignificant. So, it is possible that access to finance is capturing some part of the corruption perception, although we find the correlation coefficient between these two variables to be 0.29. And, once again we remind our readers that in the presence of interaction terms, the explanatory power of the variables is of little significance as explained above. We therefore need to estimate the conditional effects (Dutta and Sobel, Reference Dutta and Sobel2021; Berry et al., Reference Berry, Golder and Milton2012; Braumoeller, Reference Braumoeller2004). The obstacle to finance has a mean value of 1.2 (Table 1), which implies a low degree of perceived obstacle among the respondents. Secondly, it would have been useful if obstacle to finance had questions regarding perceived corruption which WBES does not include at present. This should help with qualifying the present surmises regarding how much of the obstacle to finance is owing to the perceived corruption. Third, in order to check whether the correlation can be a potential concern for bias, we consider variance inflation factor (VIF). Usually, a VIF higher than 10 for a variable warrants further investigations. Sometimes, the considered threshold is Tolerance which is 1/VIF and, thus, the critical number becomes 1/10. For our regressions in both columns (1 and 2), we check the VIFs. For both corruption perception and access to finance variables, the VIF is less than 2 suggesting that the correlation between the variables is not a cause for concern. Yet, to further mitigate concerns, we construct a dummy for finance perception variable. The dummy takes 1 for above average values of accessing finance difficulties, 0 otherwise. We interact the dummy with corruption perception variables. The results are presented in column (5) of table 3. We find that the interaction term (findummy*corr) is negative and significant similar to our results in column (3). The marginal estimates for column (5) specification are presented in column (6). As evident from the marginal estimates, as perceived corruption rises, firms experiencing above average perceived obstacles in accessing finance are about 2 percent less likely to innovate. This effect is lesser in magnitude relative to marginal estimates in column (4) as described in the preceding text. Yet, that is not surprising because above average values of the finance dummy will imply severe as well as very severe obstacles based on our sample.

Robustness analysis

Other perceived obstacles by firms

Do other perceived obstacles by firms affect the relationship between probability to innovate and corruption perception? In table 4, we proceed to explore this question. The survey asks questions to the firm owners about their perceived obstacles related to electricity, telecommunication, transportation, and customs. The questions are structured similar to the other obstacle-related questions. Each obstacle measure ranges from 0 to 4 with 0 indicating at particular obstacle not being a hindrance according to the respondents representing these operations. Again 4 represents the strongest perceived obstacle. The logit estimates for electricity, telecommunication, transportation, and customs are presented in columns (1), (2), (3), and (4) of table 4, respectively. As before, we find that the interaction term of corruption perception with the alternate perception measures is negative and significant for each of the specifications.

Table 4: Logit Specifications: Probability to Innovate by Firms, Perceived Corruption, and Other Perceived Obstacles

Note: The dependent variable is Innovation: a dummy taking the value 1 “if in the last three years, the establishment has introduced new products or services?” Corruption is our variable of interest representing corruption perceptions of firms. It is coded from 0 to 4. The question asked is How much of an obstacle is corruption? We consider different perceived obstacles of firms that are measured similarly. These are obstacles related to electricity (Electricity [perceived]), Telecoms (Telecom [perceived]), Transport (Transport [perceived]), and Customs (Customs [perceived]). Corruption is interacted with each of these perceived obstacles. We include the benchmark controls. Robust Standard Errors are in parenthesis. We control for state and industry fixed effects.

***p < 0.01; **p < 0.05; *p < 0.1.

Yet, to interpret the results properly, we need marginal estimates for our findings. Keeping space constraint in mind, we report marginal estimate for our variable of interest, ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}}}$, as noted in table 4. The marginal estimates for our control variablesFootnote ³ are available on request. The marginal estimates associated with the control variables support our benchmark findings. Here also, ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}}}$ is estimated for the minimum and maximum values of the alternate perceived obstacle variables. Consequently, the overall conclusions remain the same. $\Delta {{\delta Inno_{ijs}} \over {\delta Corr_{ijs}}}$ from the lowest obstacle to the most severe obstacle is always negative suggesting that probability to innovate is dampened for rise in the perceived corruption as firms apprehend obstacles in different forms. At the higher level of perceived obstacles, higher corruption perceptions lower probability to innovate between 5 and 7 percent.

Identification

Inverse probability weight estimates

As mentioned earlier, endogeneity can also arise out of sample selection bias if both the probability to innovate and corruption perceptions are codetermined. Under such circumstances, Borin and Mancini (Reference Borin and Mancini2016) advocate employing such estimation methods to resolve sample selection bias. Mallick and Yang (Reference Mallick and Yang2013) point out that propensity score matching reduces sample selection bias by creating a carefully matched control group (Dutta et al., Reference Dutta, Kar and Sobel2021; Webster and Piesse, Reference Webster and Piesse2018). Ideally, to resolve sample selection bias, we need to observe the same firm under two circumstances. The same firm under the two circumstances should have all attributes exactly the same except differing with regard to corruption perceptions. However, it is not possible to observe the same firm in two different scenarios. Yet, we can create a counter factual that implies having firms with all the matching characteristics except differing in terms of perceptions about difficulties in accessing finance and corruption. We consider Inverse Probability Weight (IPW) estimates that correct for missing data problems arising from the fact that each firm is only observed in one of the potential outcomes by using estimated probability weights. A two-step approach is used for estimating the treatment effects. In the first step, the parameters of the treatment model are estimated and then the estimated inverse probability weights are computed. In the second step, the weighted averages of the outcomes for each treatment level are computed using the estimated inverse probability weights. The contrasts of these weighted averages provide the estimates of the average treatment effects (ATE) (STATA, 2019).

We report the ATEs in table 5 based on IPW estimates. In column (1) of table 5, we present the average treatment effects with the treatment group being above average levels of corruption perception. For IPW estimates both the dependent and the independent variable need

Table 5: IPW and AIPW Estimates: Probability to Innovate for Firms and Perceived Corruption, and Accessing Finance Obstacles (Perceived)

Note: The dependent variable is Innovation: a dummy taking the value 1 “if in the last three years, the establishment has introduced new products or services?” The treatment group in column (1) includes firms that perceive above average corruption. In column (2), the treatment group includes firms perceiving both above average corruption and accessing finance obstacles. Finally, in column (3), the treatment group consists of firms that perceive below average corruption but above average obstacle in accessing finance (perceived). We include the benchmark controls. We control for state and industry fixed effects.

***p < 0.01; **p < 0.05; *p < 0.1.

to be binary. We find the average treatment group is negative and significant suggesting that firms perceiving above average levels of corruption are about 1.8 percent less likely to innovate.

To test whether attitudes about accessing finance difficulties of firms matter along with corruption perceptions, we create the next treatment group in column (2) of table 5. In column (2), the treatment group consists of firms perceiving above average levels of corruption as well as above average levels of difficulties in accessing finance. The average treatment effect is negative and significant. Firms perceiving both above average levels of corruption and accessing finance obstacles are about 3 percent less likely to innovate. Finally, in column (3), we consider firms with above average perceptions about accessing finance but below average levels of perceived corruption. Because our earlier findings show that corruption perceptions hinder innovation and perceived obstacles in accessing finance is critical in the presence of corruption, we want to check the same with IPW estimates. Indeed, we find that ATE is positive and significant indicating that when corruption perceptions are low, even higher perceived obstacle in accessing finance by firms does not deter their innovation efforts. Overall, our conclusions remain unaltered.

In table 5, we present the augmented inverse probability weight (AIPW) estimates. AIPW estimates, similar to IPW estimates, correct for the missing data problem arising from the fact that each firm is observed in only one of the potential outcomes. The advantage of AIPW estimates is that it uses an augmentation term in the outcome model to correct the estimator in case the treatment model is misspecified. The format in table 5 of IPW estimates is similar to that of table 5 AIPW estimates. As evident from the results, our conclusions remain unchanged.

Omitted variable bias: Additional controls

Yet another way to handle endogeneity is to control for additional factors that can also affect the probability to innovate by firms. While we have included a series of benchmark controls as described in the preceding text, in this section we control for additional variables that can mitigate omitted variable bias further. Following Pellegriono and Savona (Reference Pellegrino and Savona2017) and Bertoni and Tykvovà (Reference Bertoni and Tykvová2015), we control for square of age check for possible nonlinearity effects. Piva and Vivarelli (Reference Piva and Vivarelli2009) have shown that presence of skilled workers can boost innovation efforts as they might be adept at coming up innovative ideas (Song et al., Reference Song, Almeida and Wu2003). Our dataset does not provide breakdown of workers in a firm based on skill levels. The only variable we have at our disposal is years of education of an average worker in a firm. We control for the education variable. Because this variable results in loss of about 2,000 observations, as an alternate proxy of the education variable we control for the years of experience of the top manager. We also control for percent of losses in sales of the firm due to power outage. Percent share owned by the largest firm in the establishment is the next variable for which we control. Although not a direct proxy, it provides us with an indirect proxy for organization of the firm, Bresnahan, Brynjolfsson, and Hitt (Reference Bresnahan, Brynjolfsson and Hitt2002) have stressed the importance of organizational structure in the context of innovation.

The results are presented in appendix 1 keeping space constraint in mind. We consider our benchmark specification from table 3, column (2) as specified in equation (2). In column (1), we control for the square of age. In columns (2) and (3), we control for education of workers years of experience of the top manager simultaneously. In column (4), we control for percent loss in sales due to power outage. Finally, in column (5), percent share of the largest firm in the establishment is considered. Keeping space constraint in mind, we present the log it estimates along with marginal estimates for ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}}}$. But we do not present the marginal estimatesFootnote ⁴ for our control variables and they are available on request. We find our results remain robust. The coefficient of the interaction term is negative and significant in all the specifications and the significance level is at 1 percent for three out of the five specifications. The marginal effects support our benchmark conclusions.

Identification: Heckman selection model

Finally, we consider Heckman selection model to take account nonrandom selection and, thus, the possible bias arising out of it. It can very plausible that firms’ need for loans or not dictate their perception about corruption or financial barrier. Thus, there might be the possibility of observing a nonrandom sample. As Muravyev et al. (Reference Muravyev, Talavera and Schäfer2009) points out, firms who do not need financing because of having sufficient internal funds are more likely to get a loan. However, firms may not apply for loans at all because they perceive a higher obstacle. Thus, we estimate our model with a Heckman correction. Based on this, we can fit the regression model (y = Xβ) that consists of our dependent and independent variables. The variable(s) Z determines whether the dependent variable is observed or not. Thus, we include the need for a loan in the second list and all other independent variables of interest including perceived corruption as well as perceived obstacle in accessing finance along with the interaction term in the first list.

Keeping space constraint in mind, we do not report the findings but they are available on request. The variables from the selection list or the second list that includes the need for loan variable is available on request. The loan variable is a dummy that indicates if the firm has applied for a loan in the last fiscal year or not. Our coefficient of interest—the interaction term—is negative and significant suggesting that our main conclusions remain robust. Interestingly both perceived obstacles—corruption and access to finance—are positive and significant. But as we know in the presence of an interaction term we have to meaningfully analyze the economic significance using the total effect as given by ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}}} = {\rm \alpha }_1 + {\rm \alpha }_3Fin_{{\rm ijs}}$. Based on marginal effects, we find while ${{\delta Inno_{ijs}} \over {\delta Corr_{ijs}}}$ is positive and significant at the lowest value of perceived difficulty in accessing finance, it does turn negative and significant for the strongest financial barrier as perceived by firms. The significance of the chi-square from the Wald test of significance suggests the appropriateness of the Heckman estimates.