Linguistic Tone and Internal Control Reporting: Evidence from Municipal Management Discussion and Analysis Disclosures

In 1999, the Governmental Accounting Standards Board (GASB) introduced GASB Statement No. 34, Basic Financial Statements—and Management's Discussion and Analysis—for State and Local Governments, which requires that state and local governments include a management discussion and analysis (MD&A) section in their Comprehensive Annual Financial Report (CAFR).¹ The purpose of this provision is to provide citizens with an “objective and easily readable analysis of the government's financial performance for the year” (GASB 1999, Cod. 2200.106). While prior research has studied the financial reporting consequences of linguistic characteristics in corporate disclosures (Goel, Gangolly, Faerman, and Uzuner 2010; Loughran and McDonald 2011; Allee and DeAngelis 2015), extensions to the municipal sector are limited. The purpose of this study is to model the determinants of MD&A linguistic tone in the municipal sector and evaluate whether MD&A linguistic tone has predictive value in terms of future financial reporting via internal control quality.

Our research is motivated by the GASB's recent improvement project to explore options for enhancing the financial statement analysis section and reducing the prevalence of boilerplate language in required disclosures (GASB 2019).² Under GASB 34, the MD&A in annual financial reports is “designed to give readers a clearer picture of the social and economic factors affecting the fiscal health of the reporting body” (Guo, Fink, and Frank 2009). Since municipal managers have incentives to signal about financial reporting to reduce information asymmetries with stakeholders (Evans and Patton 1987), one potential mechanism is through textual content in the MD&A (Sacco and Busheé 2013).

Financial problems are often associated with internal control quality, and current research shows the consequences of weak internal control quality ranging from higher municipal interest costs (Baber, Gore, Rich, and Zhang 2013; Cuny 2016; C. Edmonds, J. Edmonds, B. Vermeer, and T. Vermeer 2017) to greater finance director turnover following accounting restatements (Rich and Zhang 2016). In this study, we investigate whether municipal managers signal future underlying financial problems through textual disclosures. If the language in MD&A disclosures provides useful information about financial reporting systems, one consequence could be a measurable association between the linguistic tone of the MD&A and future internal control weaknesses.

We base our empirical analyses on a sample of 362 municipal MD&A disclosures in fiscal year 2011, and we begin with a determinants model of linguistic tone calculated as the fraction of positive or negative words to total words (Loughran and McDonald 2011; Mayew, Sethuraman, and Venkatachalam 2015). Our determinants analysis suggests that contemporaneous governance, demographic, and performance characteristics impact the linguistic tone of MD&A disclosures. More specifically, municipalities with higher unemployment rates, lower education levels, general fund deficits, less intergovernmental revenue, and current internal control weaknesses are associated with greater use of negative language, while the council-manager form of government is associated with the use of more positive language. These findings suggest that MD&A linguistic tone captures information about a municipality's financial environment.

The results from our future internal control analysis suggest that positive (negative) linguistic tone in municipal MD&A disclosures is associated with fewer (more) subsequent-year internal control weaknesses after controlling for other governance, demographic, and performance factors. To the extent that internal controls impact financial reporting systems in general, our findings imply that study of the word choices made by municipal managers in MD&A disclosures is useful in forecasting the quality of future financial reporting for local governments. Our results on linguistic tone and future internal control weaknesses could also suggest future financial problems for the municipality. In a supplementary analysis, we find that municipalities with more positive tone are less likely to experience future general fund deficits, suggesting that linguistic tone does have predictive value regarding financial health.

We contribute to the accounting literature in several ways. First, while many papers examine the linguistic tone of corporate MD&A disclosures (Loughran and McDonald 2011; Lehavy, Li, and Merkley 2011; Brown and Tucker 2011; Davis and Tama-Sweet 2012; Mayew et al. 2015, among others), we know relatively little about the role of qualitative disclosure information in the governmental setting (Rich, Roberts, and Zhang 2016; Sacco and Busheé 2013; Guo et al. 2009; Marsh, Montondon, and Kemp 2005). Our paper utilizes automated linguistic techniques (Loughran and McDonald 2016; Fisher, Garnsey, Goel, and Tam 2010) to extend current research of MD&A disclosures to the municipal context. Second, our results on the determinants of linguistic tone in municipal MD&A disclosures enhance understanding of the disclosure decisions made by local government managers, particularly with respect to qualitative, nonfinancial information. Third, our robustness checks inform the scenarios (such as current report delay) that influence the predictive value of positive and negative language.

Fourth, our study informs on the effectiveness of required MD&A disclosures under GASB 34 for local governments and the benefits of the GASB's current project to reexamine the financial reporting model, including disclosures required in the MD&A section (GASB 2019). Fifth, our empirical findings add to the current literature on the factors associated with internal control quality in the local government. This is important in the governmental sector, given that studies have shown that there are consequences to poor quality (Baber et al. 2013; Cuny 2016; Edmonds et al. 2017; Rich and Zhang 2016). The results of our study will therefore benefit citizen stakeholders in evaluating the stewardship role of government officials (GASB 2006). Finally, following corporate textual analysis literature that finds linguistic tone is associated with future earnings and return on assets (e.g., Li 2010; Davis, Piger, and Sedor 2012), we analyze linguistic tone's impact on municipal financial performance. Our results that linguistic tone is associated with financial performance in the municipal government aids future researchers in this area in terms of managers' signaling behaviors.

We organize the remainder of this study as follows. First, we present a literature review that includes a background on the use of textual analysis in evaluating disclosures, followed by a review of the literature on internal control quality, and a discussion of our hypothesis. Our research design, including sample selection procedures and methodology, is presented next, followed by a discussion of the results and robustness checks. The last section provides concluding remarks.

Corporate research suggests that the linguistic tone in management disclosures helps explain both accounting and market returns (Loughran and McDonald 2011; Li 2010; Feldman, Govindaraj, Livnat, and Segal 2010; Henry 2006), signifying that qualitative disclosures contain information that is incrementally valuable to investors. Additionally, tone is shown to influence shareholder voting results regarding executive compensation (Balsam, Boone, Liu, and Yin 2016), consistent with stakeholders viewing linguistic tone as a proxy for managerial performance. Furthermore, prior evidence links linguistic tone to going-concern and bankruptcy filings (Mayew et al. 2015; Shirata, Takeuchi, Ogino, and Watanabe 2011; Shirata and Sakagami 2008) and fraud (Loughran and McDonald 2011; Goel et al. 2010). These latter findings imply that managers make deliberate language choices when facing difficult firm conditions.

Related to our study, current research shows that the textual content of management disclosures contains value-relevant information in explaining internal control quality in the corporate sector. By developing a new word list designed to more accurately measure tone in financial text, Loughran and McDonald (2011) demonstrate that tone, calculated using their new word list, is related to internal control material weaknesses during their sample period from August 2002 to November 2005. Specifically, they find that firms disclosing a material weakness are more likely to use negative language in their 10-Ks than those without internal control problems.³ Overall, their results suggest that MD&A tone can play an important role in identifying firms with internal control problems.

However, MD&A disclosure can be uninformative if it does not change with economic changes at the firm (Brown and Tucker 2011). Brown and Tucker (2011) find empirical evidence that while MD&A disclosures became longer during 1997–2006, they consisted of increasingly boilerplate language that is less useful in explaining stock returns. Moreover, the SEC (1998) has expressed some concerns over the use of boilerplate language in corporate MD&A disclosures. In addition, the GASB's (2019) recent project to reexamine GASB 34 sets out to explore the elimination of requirements in the MD&A section that are boilerplate and no longer necessary for understanding the financial reporting model. Thus, MD&A disclosures in the municipal government may also consist of stale text that has little information content in predicting future financial reporting quality.

Although current research shows that GASB No. 34 has information content in the debt market (Callahan and Waymire 2015; Reck and Wilson 2014; Pridgen and Wilder 2013; Plummer, Hutchison, and Patton 2007), research also finds variation and lack of detail in disclosure under GASB 34 for some local governments (Vermeer, Patton, and Styles 2011). Based on GASB 34 (GASB 1999, Cod. 2200.109c) the MD&A “should provide users with the information they need to help them assess whether the government's financial position has improved or deteriorated as a result of the year's operations.” Although prior literature suggests that textual content in municipal MD&A disclosures is useful in signaling recessionary times (Sacco and Busheé 2013), there is little empirical evidence on the consequences of linguistic characteristics within municipal MD&A disclosures.

Several studies suggest that local government disclosures need to be more readable for citizens. For example, Kinnersley and Fleischman (2001) compare the readability of state and local government transmittal letters (TLs) to the MD&As of publicly traded companies. They find that government TLs tend to be more difficult to read than publicly traded firm MD&As based on readability scores driven by word and sentence length. Due to the complexity of words, they also suggest that a college reading level is required to understanding both documents. Clarke, Hrasky, and Tan (2009) examine whether narrative characteristics vary between the reports of local governments and listed companies. They find that mayoral letters show greater reading ease and have less obfuscation than the respective corporate letters. In addition, mayoral letters tend to be shorter and contain more passive constructions. In examining the linguistic tone of municipal disclosures, Rich et al. (2016) find that MD&A linguistic tone has information content in predicting future financial reporting delays. They find that the proportion of positive to total words in MD&A disclosures is related to more timely financial reporting in the following year after controlling for current report timing and other municipality and governance factors.

The federal government provides annual grants in excess of $400 billion to state and local governments and other nonprofit organizations (https://obamawhitehouse.archives.gov/omb/financial_fin_single_audit/). Entities that spend $750,000 or more of federal funds in a year ($500,000 for fiscal years ending on or before December 31, 2014) are subject to the audit and internal control requirements of Circular A-133, also known as single audits (OMB 2014).⁴ The objectives of an A-133 audit are to provide assurance that the entities receiving federal grants comply with laws, regulations, and the provisions of contract and grant agreements, and to maintain internal control to provide reasonable assurance of compliance with these requirements (OMB 2003). Internal control systems are an important input to financial reporting quality since they help to predict or prevent future significant accounting and financial reporting problems. In addition, internal control quality is important for measuring the stewardship role of the municipal officials and managers for efficient use of taxpayer money and holding them accountable in utilizing public resources (GASB 2006).

Many studies examine internal control quality in the governmental setting (Wallace 1981; Ziegenfuss 2001; Boyle, Cooper, and Geiger 2004; Rich and Zhang 2014; Johnson, Hartong, and Kidd 2014). For example, Wallace (1981) analyzes the internal control reporting practices within municipalities. The results suggest that existing disclosures fail to properly address risk exposure, costs, and benefits. In examining the consequences of internal control quality, Ziegenfuss (2001) investigates whether a local government's control environment is related to the incidence of fraud. Based on a survey of governmental internal auditors, results show that the strength of a local government's control environment is negatively associated with incidence of fraud. Moreover, Gore, Henderson, and Ji (2016) examine whether the presence of internal control problems impacts municipal bond markups. They find that municipal bonds issued by entities with a material weakness in internal controls exhibit larger markups. Their evidence suggests markups reflecting compensation to underwriters for the additional risk and effort involved in pricing bonds issued by entities with serious accounting problems.

In examining the determinants of internal control quality, prior research finds that the presence of internal control weaknesses is positively associated with counties rather than cities (Jakubowski 1995), public CPA firms rather than governmental auditors (López and Peters 2010), and lengthier audits (Modlin 2012), while being inversely associated with cities holding the Government Finance Officers Association's (GFOA) Certificate of Achievement for Excellence in Financial Reporting (COE) (Cox and Wichmann 1993). Specifically, Jakubowski (1995) investigates auditor reports about internal control structures of local governments and how the structures changed over the first four reporting years under the Single Audit Act of 1984 (SAA). His results show that auditors report more control weaknesses in counties than in cities, and that the number of control weaknesses reported declined during the first four reporting years under the SAA for city governments.

Using survey opinions of government managers, Cox and Wichmann (1993) find that internal controls were perceived as stronger for cities holding the COE and having “other” special districts (e.g., enterprise funds) than other units of government. López and Peters (2010) examine the relationship between auditor type and the likelihood of auditor-disclosed internal control concerns using a sample of single audit reports from U.S. cities and counties. They find that CPA firms are more likely to issue internal control concerns in their audit reports than governmental auditors, and that the difference in internal control reporting is even more noticeable for reports issued by larger CPA firms.

Modlin (2012) investigates independent auditor findings for control systems within North Carolina county governments. The results show that a majority of county governments have multiple reporting issues that range from internal control problems to reconciliation issues. The likelihood of reporting problems increases when governments have lengthier audits, less expensive audits, or a limited number of accountants or accounting specialists.

Elder, Kattelus, and Ward (1995) compare finance officer and auditor assessments of internal control for a sample of Michigan municipalities. They find that finance officer assessments were more favorable of their control systems than the auditor assessments were for the same systems. In addition, auditors' internal control assessments were significantly more associated with the number of audit adjustments than the finance officers' assessments were. Our paper contributes to this literature by examining the information content of municipal managers' disclosures in forecasting future internal control quality.

Similar to corporate sector findings linking linguistic tone and internal controls (Loughran and McDonald 2011), language in municipal MD&A disclosures could also provide information to financial statement users about future internal control quality. Although MD&A sections are required under GASB 34, whether the textual characteristics in the MD&A have information content related to financial reporting remains largely an empirical question. On one hand, existing research suggests that municipal officials and managers have incentives to signal on financial reporting in order to reduce information asymmetries with various stakeholders (Baber 1983; Evans and Patton 1987). A potential mechanism for signaling performance is through the textual content in municipal MD&A disclosures (Sacco and Busheé 2013). If MD&A language contains useful information about financial reporting systems, one possibility would be an observation of associations between MD&A linguistic tone and future internal control quality.

On the other hand, governmental financial reports are deemed significantly complex and unfamiliar to some stakeholders (Mead 2002). Thereby, the usefulness of incremental MD&A information content may be limited. Furthermore, given that job security is a significant factor in deciding to work for governmental organizations (Lewis and Frank 2002), managers may have few incentives to signal performance since it may have little effect on career prospects. Finally, there is the chance that qualitative MD&A sections are primarily “boilerplate” in nature with limited information content (Brown and Tucker 2011). As a result, we may not observe an association between MD&A tone and future internal control weaknesses. Given these opposing possibilities, we present the following hypothesis (stated in null form).

Similar to recent work (Rich et al. 2016), we begin with the 3,566 municipalities that responded to a 2011 Municipal Form of Government Survey conducted by the International City/County Management Association (ICMA). In the interest of facilitating the data acquisition process, we draw our sample from the 679 responding municipalities with a population greater than 25,000 citizens. We then identify municipalities with audit data for both 2011 and 2012 in the Single Audit Database (SAD) available from the Federal Audit Clearinghouse (https://harvester.census.gov/facweb/Default.aspx) to allow for the collection of internal control reporting and other audit details. Next, we use a web-based search of municipality websites to obtain CAFRs for 2011, and extract the MD&A section for textual analysis.⁵ Following prior studies (Mayew et al. 2015; Rich et al. 2016), we import the Loughran and McDonald (2011) dictionary into the Linguistic Inquiry and Word Count (LIWC) software to calculate linguistic “tone” based on the fraction of positive and negative words to total words in the MD&A.⁶

Data on governance factors are from the ICMA survey, while details on income and educational achievement at the county level are from the American Community Survey (ACS) administered by the U.S. Census Bureau.⁷ County-level unemployment statistics are from the Local Area Unemployment Statistics provided by the Bureau of Labor Statistics (BLS), and we obtain the list of recipients of the Government Finance Officers Association's (GFOA) Certificate of Excellence in Financial Reporting directly from the GFOA. After deleting municipalities with missing audit details, MD&A disclosures, and governance or demographic data, the final sample consists of 362 unique municipalities. Table 1 outlines our sample selection procedures.

We first perform an analysis to provide insight on the determinants of linguistic tone in municipal MD&A disclosures.⁸ More specifically, we model current linguistic tone as a function of current governance, demographic, and performance factors as follows:

Similar to prior work (Mayew et al. 2015; Rich et al. 2016), the dependent variable (TONE) is measured using proportions of total words in MD&A disclosures. We utilize measures capturing the percentage of words that are (1) positive, (2) negative, and (3) net positive (i.e., positive − negative).

Our independent variables are defined as follows, with predictions based on the likelihood of positive linguistic tone.⁹ First, prior literature suggests that the council-manager form of government can positively impact financial reporting decisions (Evans and Patton 1983), possibly because of increased incentives for council members to monitor the chief executive. Therefore, we include CNCL_MGR as an indicator variable equal to 1 for municipalities where the city manager is the chief executive, and 0 otherwise, but we make no prediction in our model regarding linguistic tone. Next, results from Baber et al. (2013) show that municipalities with staggered council elections (and possibly more entrenched elected officials) are more likely to restate. As a result, we include C_STAG as an indicator variable equal to 1 for municipalities with staggered council election cycles, but do not make any directional predictions in our determinants model.

We include two variables designed to capture auditor expertise. First, we include the fraction of A-133 municipal federal funds audited by a particular auditor in a given state (AUD_MKT_SHR) to consider the impact of specialization in municipal audits. Additionally, we include BIG4 as an indicator variable equal to 1 for audits conducted by a member of the Big 4 audit firms, and 0 otherwise. We predict that both of these proxies will be associated with managerial confidence in future audit results, which should lead to more optimistic MD&A language and a positive coefficient for each in our model. Additionally, based on the characteristics of the municipality, engagements are classified as “low risk” as part of the A-133 audit process, implying that the auditors expect fewer exceptions to arise during fieldwork (López and Peters 2010; Petrovits, Shakespeare, and Shih 2011). We include LRISK as an indicator variable equal to 1 for low-risk audits, and predict that lower-risk municipalities will have healthier financial reporting structures and therefore will be more likely to use positive language in MD&A disclosures. We also include the number of federal agencies (AGENCIES) providing funding to a given municipality to consider additional outside monitoring, but do not make a prediction on the direction in our model in terms of linguistic tone.

We consider the following demographic factors in our model; we begin with municipality size through the logged number of citizens (POPLTN). Furthermore, we consider the impact of administrative complexities necessary to manage federal programs via the log of total federal funds spent (FEDFUNDS). Given an unclear relationship between these proxies and linguistic tone, we make no predictions in our model. Next, we control for jurisdiction wealth via per capita income (PC_INCME) and education through the fraction of citizens with at least a bachelor's degree (PCT_BCHLR). We predict that a wealthier and more educated population indicates a more stable funding base, which will be directly linked to more optimistic MD&A language and a positive coefficient in our model. Finally, we include the unemployment rate (UE_RATE) to consider the economic conditions present in a municipality. We predict that higher unemployment rates will be associated with a less optimistic tone in disclosures, resulting in a negative coefficient in the model.

We include two fiscal performance measures in our model. GF_ADJ_DEF is an indicator variable equal to 1 to denote cities with current general fund deficits after adjusting for other financing sources/uses, and 0 otherwise. Given that deficits likely signify financial distress in that they reflect outflows greater than inflows, we predict a less optimistic tone and a negative coefficient in our model. Furthermore, we include GF_IG_REV as the fraction of intergovernmental to total general fund revenue (i.e., the percentage of funding derived from higher levels of government). If greater levels of intergovernmental revenue represent less fiscal self-reliance (Berry and Gersen 2009; Zhang and Rich 2016), one consequence could be uncertainty about future funding levels. However, obtaining revenue from the state and federal government to fund the municipality may suggest the ability to subsidize municipal services with outside sources, therefore we make no prediction in our model.

We consider three accounting performance metrics as follows. First, we include GFOA_COE as an indicator variable equal to 1 to denote cities receiving the GFOA's Certificate of Achievement for Excellence in Financial Reporting Program (COE), and 0 otherwise. Participation in the COE program is voluntary and signifies a commitment to meeting certain financial reporting best practices, so we predict this will relate to optimistic language and a positive coefficient in our model. Next, we include RDELAY as the logged number of days between the fiscal year-end and the date the auditor signs the financial statements. Rich et al. (2016) find that positive tone is inversely associated with report delay, so we predict a negative coefficient in our model. Finally, we include an internal control index (ICW_INDEX) that takes on a value of 2 for material weaknesses, 1 for significant deficiencies, and 0 for observations with no disclosed internal control weaknesses (López and Peters 2010). As a result, high values of ICW_INDEX are likely an indication of poor-quality internal controls. Given that disclosed internal control weaknesses likely indicate problems in the financial reporting system, we predict this will lead to uncertainty about financial reporting that results in a negative coefficient in our model.

Our primary goal in this study is to test whether the linguistic tone of current MD&A disclosures has predictive value in terms of future internal control weaknesses. As such, we construct Equation (2) as follows:

The dependent variable for our ordered logit specification, F_ICW_INDEX, is calculated using the same parameters as for the current year ICW_INDEX described above, and is designed to capture the existence and severity of future internal control weaknesses. However, it is measured using 2012 data, which is one year ahead of all independent variables. Once again, high values of F_ICW_INDEX reflect lower quality internal control outcomes. The variables of interest are based on the three TONE variables (positive, negative, and net positive) from Equation (1). Consistent with Equation (1), we include other explanatory variables that capture current governance, demographic, and performance factors, with all values measured using 2011 data. We include state fixed effects to control for other regulatory differences that could impact disclosure decisions. Values for independent continuous variables are winsorized at the 1 percent and 99 percent tails of the distribution to minimize the effect of influential observations, and standard errors are clustered on states (Rogers 1993).

Table 2 provides the descriptive statistics for the variables used in our analysis, with Panel A summarizing the entire sample. The use of positive and negative words is consistent across the overall sample, as highlighted by the mean values of 0.48 for POS_TONE, 0.49 for NEG_TONE, and 0.00 for NPOS_TONE. The incidence of internal control weaknesses appears to be relatively consistent over our two-year sample period, as highlighted by the value of 0.55 for ICW_INDEX combined with the average future internal control index (F_ICW_INDEX) of 0.54.¹⁰

Table 2, Panel B presents statistics for the sample divided into two categories based on the presence of at least one internal control significant deficiency or material weakness in year 2012 (F_ICW_INDEX > 0), along with tests of mean differences between the two subsamples. The results highlight that municipalities with future internal control weaknesses are statistically more likely to use negative tone (NEG_TONE) in their current MD&A disclosures than municipalities without internal control weaknesses, which is preliminary evidence consistent with our hypothesis. Future internal control weaknesses are strongly associated with current internal control weaknesses and report delay, which is consistent with prior literature (Petrovits et al. 2011; Modlin 2012). In terms of other control variables, future internal control weaknesses vary directly with current size (both in terms of the number citizens and the amount of federal funds audited), the number of federal agencies providing funding, intergovernmental revenue, unemployment rate, and a Big 4 auditor, and inversely with low-risk audit designations, receipt of the GFOA's financial reporting certificate, percentages of citizens with a bachelor's degree, and per capita income.

Table 3 provides a correlation matrix for the variables used in our analysis. The correlations corroborate some of the findings from Table 2, in that they highlight a direct correlation between future internal control weaknesses (F_ICW_INDEX) and negative linguistic tone (NEG_TONE). Alternatively, the results suggest no statistically significant correlation between future internal control weaknesses and positive linguistic tone (POS_TONE). We also note a significant and direct correlation between future and current internal control weaknesses, consistent with prior studies (Petrovits et al. 2011). We also note that the only other correlations among independent variables that exceed 0.40 are those between per capita income levels (PC_INCME) and educational attainment (PCT_BCHLR), and federal funds (FEDFUNDS) and population (POPLTN). We discuss statistical tests for the influence of multicollinearity in our “Robustness Tests and Other Analyses” section.

Table 4 presents the OLS results from our determinants model of linguistic tone based on Equation (1), with all variables measured using 2011 values. Column 1 highlights that only the council-manager form of government is statistically associated with the use of positive tone (CNCL_MGR = 0.05; z-statistic = 2.25), which is consistent with greater optimism when the municipality is run by a professional manager. The lack of significant explanatory variables is consistent with assertions in Loughran and McDonald (2016) that positive tone can be problematic due to the use of positive words in framing negative statements (e.g., “increased shortfall”). In Column 2, results show that negative tone is directly associated with higher unemployment rates (UE_RATE = 1.71; z-statistic = 3.14), general fund deficits (GF_ADJ_DEF = 0.05; z-statistic = 2.24), and internal control weaknesses (ICW_INDEX = 0.08; z-statistic = 2.56). Given that each of these proxies represents a challenging municipal condition, we interpret these results as evidence that the linguistic tone of municipal MD&A disclosures captures information about the well-being of a given local government. Furthermore, we find that localities with less educated populations are more likely to use negative language (PCT_BCHLR = −0.25; z-statistic = −1.76), possibly suggesting a lack of confidence in the tax base. We also show that greater levels of intergovernmental revenue are associated with less negative linguistic tone (GF_IG_REV = −0.21; z-statistic = −2.54), possibly because of confidence in financial condition being bolstered by higher levels of government.

The results in Column 3 for net positive tone are largely consistent with those in columns 1 and 2, although there is also a significantly negative coefficient on current report delay (RDELAY = −0.08; z-statistic = −1.87), which is consistent with report delay being perceived unfavorably. Overall, we interpret our results to be consistent with language choices in MD&A disclosures containing important information about a municipality's condition to stakeholders.

Table 5 presents the multivariate results for ordered logit specifications based on Equation (2). Recall that future internal control weaknesses are measured using 2012 data, while all independent variables are measured using 2011 data.¹¹ Columns 1 and 2 represent base conditions, where we do not include TONE measures and we vary the inclusion of current internal control weaknesses. The baseline results in Column 1 highlight that future internal control weaknesses are positively associated with the number of agencies providing federal funding (AGENCIES = 0.39; z-statistic = 2.55), federal funds spent by the municipality (FEDFUNDS = 0.28; z-statistic = 1.66), and audit report delay (RDELAY = 1.64; z-statistic = 3.47), while being negatively associated with the low-risk audit designation (LRISK = −0.63; z-statistic = −3.54) and GFOA certificate (GFOA_COE = −0.56; z-statistic = −1.88).

In Table 5, Column 2, we show that future internal control weaknesses are strongly correlated with current internal control weaknesses (ICW_INDEX = 1.72; z-statistic = 7.00), and that the information content of ICW_INDEX appears to supersede that of LRISK and AGENCIES. In terms of performance factors, we find that future internal control weaknesses are positively associated with greater levels of intergovernmental revenue (GF_IG_REV = 1.78; z-statistic = 2.15), suggesting that municipalities requiring support from higher levels of government are more likely to have future internal control problems. Furthermore, consistent with results from Cox and Wichmann (1993), participation in the GFOA's financial reporting certificate program is negatively associated with future internal control problems (GFOA_COE = −0.82; z-statistic = −3.18), suggesting that the requirements for participation in the program have a positive impact on internal control structures. Finally, consistent with Modlin (2012), we find direct links between future internal control weaknesses and current report delays (RDELAY = 0.82; z-statistic = 1.96), indicating that reporting delays signal problems in financial reporting systems.

The results in Table 5, Columns 3 through 5 include our TONE proxies into a series of specifications that vary the inclusion of current internal control weaknesses (ICW_INDEX) and state fixed effects. The findings consistently show that current linguistic tone has predictive value in terms of future internal control weaknesses. As hypothesized, our analysis shows that higher values of our future internal control index are associated with less positive and more negative language, as highlighted by the negative coefficients on POS_TONE (coefficient = −2.14; z-statistic = −3.16), and the positive coefficient on NEG_TONE (coefficient = 2.07; z-statistic = 2.73) in Column 3. These findings are consistent when we control for current internal control weaknesses (ICW_INDEX) and add in state fixed effects in Columns 4 and 5, respectively. We interpret these findings as evidence that language in MD&A disclosures can help stakeholders anticipate internal control issues in the near term.

We perform a series of tests to validate the robustness of our results. First, we use the net of positive minus negative word use as a measure of linguistic tone. Results (untabulated) using this net positive tone measure are very similar to those presented in Table 5. Next, we use some alternative definitions of our independent variables, including measuring general fund deficits (GF_DEF) before other sources/uses, as well as auditor market share (AUD_MKT_SHR) based on the number of clients audited in a particular state as opposed to the magnitude of federal funds audited. Results (untabulated) with these alternative variable definitions are virtually identical to those presented in Table 5. We also manually remove all tables, charts, and running headers from our sample of MD&As and recalculate our measures of linguistic tone. While the incidence of positive and negative language increases slightly after removing these quantitative details (0.65 and 0.56, respectively), the results from our multivariate analyses using these alternative tone measures are very similar to those presented in Tables 4 and 5. Next, we review variance inflation factors for an OLS specification of our primary models (untabulated), and note no values above 5. We take this as evidence that multicollinearity is not a primary concern for our results (Kennedy 2003).

We also consider the possibility that municipal managers provide commentary on internal control structures in the text of their MD&A. We searched through each of the MD&A disclosures in our sample and identified observations that involved specific mention of strong internal controls. While we only identified four observations with this specific language, we note that they have significantly higher levels of positive language (0.77 versus 0.48; z-statistic = 3.61) than the overall sample. Inclusion of a separate indicator variable denoting these specific observations leads to results (untabulated) virtually identical to those presented in Tables 4 and 5.

In order to address concerns regarding the appropriateness of the Loughran and McDonald (2011) financial dictionary in a governmental reporting context, we compiled an extra word list for the supplemental analysis that is municipal specific. To construct our list, we use the WordStat program within the QDA Miner software to analyze the textual content of our sample of MD&A disclosures. Following Loughran and McDonald (2011), we focus on words that are present in at least 5 percent of the sample MD&As. Each word from this list is then rated as having a positive, negative, or neutral tone. All authors reviewed this dictionary to establish agreement regarding the appropriate tone of each word.

We identified 17 additional words found to have a non-neutral tone above those present in the Loughran and McDonald (2011) dictionary (see Appendix B). Specifically, our list includes 14 positive words and three negative words. As there are a greater number of positive words added, this indicates that managers utilize certain words to indicate financial strength beyond that of a corporate financial setting. When governmental MD&As include such words as awarded, compliance, surplus, and upgrades, local managers may be indicating an increase in financial condition or recognition of higher reporting quality. Thereby, these additional words could provide incremental information content within governmental reporting.

We supplement the Loughran and McDonald (2011) list with our additional words and recalculate values for positive, negative, and net positive linguistic tone. We note that the fraction of positive words increases from 0.48 to 0.58, while the fraction of negative words is virtually unchanged from the results presented in Table 2. This provides some evidence that municipality-specific words tend to be positive. However, our multivariate results (untabulated) using the updated tone proxies remain essentially unchanged to those presented in Tables 4 and 5.

We also separately examine municipalities that experienced increases versus decreases in the internal control index between 2011 and 2012 and summarize the results in Table 6. Panel A provides count statistics of the changes, where negative values reflect reductions in internal control severity and positive values reflect new or more severe internal control exceptions in 2012 compared to 2011.¹² The findings suggest that approximately the same fraction of municipalities saw increases (12 percent) and decreases (13 percent) in the internal control index for 2012, with a large majority experiencing no year-over-year change in internal control reporting (75 percent).

Table 6, Panel B provides logit regression results based on Equation (2), where the dependent variable is an indicator variable equal to 1 for increases (Column 1) and decreases (Column 2) in the internal control index discussed above. The results presented in Column 1 suggest that municipalities that experience a future increase in the internal control index utilize less positive language (POS_TONE = −2.05; z-statistic = −1.97) and more negative language (NEG_TONE = 1.45; z-statistic = 2.62) in their current MD&A disclosure, providing further support for the conclusion that current linguistic tone is informative in predicting future internal control quality. Column 2 provides similar takeaways, where municipalities that experience a future reduction in the internal control index have more positive tone (POS_TONE = 2.87; z-statistic = 2.06), although the coefficient on NEG_TONE is not significant. In terms of control variables, the results in Column 2 also suggest that wealthier municipalities and those committed to excellence in government financial reporting are more likely to reduce internal control problems, as highlighted by the positive and significant coefficients for income (PC_INCME = 2.35; z-statistic = 1.85) and receipt of the GFOA certificate (GFOA_COE = 2.19; z-statistic = 2.76).

We also test for the impact of “new” versus “continuing” internal control weaknesses in the future period. More specifically, we identify new internal control weaknesses through an indicator variable equal to 1 where F_ICW_INDEX > 0 and ICW_INDEX = 0, and continuing internal control weaknesses where F_ICW_INDEX > 0 and ICW_INDEX > 0, and 0 otherwise. We then compare each group to municipalities that had no internal control weaknesses in either year to provide a meaningful comparison group in a regression based on Equation (2).¹³ The results of this additional analysis (untabulated) suggest that our results are consistent across the two subsections, with negative tone being positively associated with both new and continuing internal control weaknesses and positive tone being inversely associated with only continuing internal control weaknesses. We interpret this finding as consistent with Table 6, Panel B.

To consider the different types of disclosed internal control weaknesses, we create indicator variables equal to 1 that singularly identify (1) significant deficiencies (SIG_DEF), (2) material weaknesses (MW), and (3) any internal control weakness (ANY_ICW) in the current and future periods, and 0 otherwise. We substitute in each of these indicators for F_ICW_INDEX and ICW_INDEX in Equation (2), and reestimate the model. The evidence presented in Table 7 (control variable estimates omitted for brevity) suggests that our findings are driven by material weaknesses, as highlighted by the insignificant coefficients on the tone measures in Columns 1 through 3 based on significant deficiencies. The results imply links between linguistic tone and the more severe material weaknesses in Columns 4 through 6, but are potentially limited by power in the state fixed effects specification. In Columns 7 through 9, we see evidence most consistent with that in Table 5, using an indicator that captures the existence of any internal control weakness. Note that the results using the any internal control weakness (ANY_ICW) measure are stronger than the results in Table 5. These findings imply that positive (negative) tone is associated with fewer (more) internal control weaknesses, which add robustness to our main analysis and indicates that our results are driven by more severe internal control problems.

Internal control weaknesses and reporting delay are potential symptoms of underlying financial problems for the municipality. We next partition the sample into two groups based on median report delay to determine if our results are driven by correlations between financial reporting system efficiency (report delay) and effectiveness (internal control weaknesses). The results in Column 1 of Table 8 for the low report delay subsample suggest that optimistic language is informative for municipalities with efficient financial reporting systems, as highlighted by the negative coefficient on POS_TONE (coefficient = −2.77; z-statistic = −2.55). One interpretation of this finding is that positive language is more credible when there is a better functioning financial reporting system. Alternatively, the results for the high report delay subsample in Column 2 imply that negative language is predictive for less efficient financial reporting systems, as highlighted by the positive coefficient on NEG_TONE (coefficient = 1.15; z-statistic = 1.96). This finding could suggest that negative language carries more weight when the financial reporting system is less robust. Taken together, we interpret our evidence to suggest that language needs to be considered in the context of the timing of financial reporting.

Next, we re-perform our primary analysis with a focus on fiscal performance instead of internal control quality. Specifically, we model the likelihood of a future general fund deficit before other sources/uses (i.e., revenues < expenditures) as a function of the governance, demographic, and performance factors (including a current general fund deficit) based on Equation (2). We present the results from this logit specification in Table 9, which provide some evidence that positive linguistic tone is associated with a reduced likelihood of future deficits. The results are consistent for a baseline model (Column 1) and one with state fixed effects (Column 2). We interpret this result to suggest that optimistic language is associated with stronger future performance, which is consistent with the findings from Rich et al. (2016).

Finally, we model logged future levels of general fund revenues (expenses) as a function of the governance, demographic, and performance factors based on Equation (2) along with logged current general fund revenues (expenses). Our results (untabulated) suggest that future general fund expenses are inversely associated with negative tone, implying that financial managers use less optimistic language when they anticipate fewer services being offered to citizens. Overall, the takeaways from these fiscal analyses is that linguistic tone is also informative regarding future fiscal performance.

In this study, we investigate the determinants and predictive value of linguistic tone in municipal textual disclosures. Using a sample of 362 municipal MD&A disclosures in 2011, we find that MD&A linguistic tone is directly associated with the council-manager form of government, education levels, and intergovernmental revenue, while inversely associated with general fund deficits, reporting delay, and internal control weaknesses. We take these findings as evidence that municipality conditions contribute to the textual choices made by managers in MD&A disclosures. Then, we present evidence that future internal control quality is associated with current MD&A disclosure tone. Specifically, our analysis shows that the fraction of positive (negative) words in municipal MD&A disclosures in 2011 is associated with less (more) internal control weakness reporting in 2012 after controlling for other governance, demographic, and performance factors. We interpret our results to suggest that municipal managers use deliberate language in MD&A disclosures based on the governance, demographic, and performance factors facing a particular locality. To the extent that internal control systems impact financial reporting outcomes, our analysis suggests that the linguistic tone in MD&A disclosures required under GASB 34 can have an important role in predicting future financial reporting quality. Moreover, our analysis also indicates that linguistic tone provides valuable information for future financial performance, such as the likelihood of a general fund deficit.

Our study is subject to certain limitations. First, the analysis is confined to municipal MD&A disclosures for the year 2011 for a sample of municipalities that both responded to the 2011 ICMA Municipal Form of Government Survey and spent enough federal funds to be subject to the Single Audit Act (OMB 2003). Thus, our results may not generalize to all municipalities. Second, our study examines only internal control quality at the local level; future research could examine future internal control and MD&A tone at the state level. Furthermore, calculation of tone proxies using the fraction of positive or negative words may not properly capture nuances in specific disclosures. Nonetheless, our research provides an incremental step toward understanding the consequences of language choices in public sector disclosures.

Overall, this study contributes to the literature in the following ways. First, while many papers examine the linguistic tone of the MD&A in corporate sector research, we know relatively little about the role of qualitative information in the governmental setting (Rich et al. 2016; Guo et al. 2009; Marsh et al. 2005). Thus, our paper extends current research using the linguistic tone of MD&A disclosures in the municipal context using automated linguistic techniques (Fisher et al. 2010). Second, our results on the determinants of the linguistic tone of municipal MD&A disclosures provide insight on the disclosure behaviors of local government managers. Third, our study informs on the effectiveness of required MD&A disclosures under GASB 34 for local governments, and the benefits of the GASB's current reporting reexamination project (GASB 2019). Fourth, our empirical findings add to the current literature on internal control quality in the local government by predicting municipal internal control quality using linguistic tone. Study of internal control quality is important because it helps stakeholders assess whether municipal officials have fulfilled their stewardship responsibilities (GASB 2006). Finally, we contribute to the literature on fiscal performance by providing evidence on the association between linguistic tone and municipal financial performance.