«Internal and External Attributions by Managers in Earnings Conference Calls by Zhenhua Chen Business Administration Duke University Date:_ Approved: ...»
I obtain accounting information from COMPUSTAT and stock return data from CRSP. I also obtain CEO characteristics from Execomp and BoardEx, and analyst forecasts from I/B/E/S in additional empirical analysis. The final sample contains 50,434 quarterly Occasionally there is more than one CEO in a conference call, often during CEO turnover periods. I drop these firms from my analysis, but keeping these firms does not change the inferences.
earnings conference calls, for which firms’ earnings and earnings forecasts are available.
The sample construction process is described in panel A of Table 1.
This table reports descriptive statistics of firm characteristics for a sample of 50,434 observations from quarterly earnings conference calls occurring between 2002 and 2007. Panel A describes sample selection process. Panel B reports descriptive statistics of firm characteristics.
Panel C presents industry composition of the sample. Variables are defined in Appendix A. All continuous variables are winsorized at the 1st and 99th percentiles.
Descriptive statistics of firm characteristics are presented in Panel B of Table 1.
The average firm has total assets of $4.8 billion, an average quarterly return on assets (ROA) of 1%, and book-to-market ratio of 0.48. The sample contains relatively large firms, consistent with prior findings that larger firms are more likely to hold earnings conference calls (Frankel et al. 1999). Prior studies find that qualitative information is also related to firm performance (Davis et al. 2008; Demers and Vega 2008; Engelberg 2008; Loughran and McDonald 2011), so I also include the frequency of positive and negative words, as defined in Loughran and McDonald (2011). The means of POSWORDS and NEGWORDS are 1.85 and 0.84, respectively. Lougrhan and McDonald (2011) report that the means of POSWORDS and NETWORDS for the MD&A portions of 10-K documents are 0.83 and 1.51, respectively. This inconsistency suggests that managers use more optimistic language during conference calls than the MD&A, which may result from managers “pushing forward” more favorable language to a more salient disclosure outlet (Davis and Tama-Sweet 2012). The fact that managers use more positive words than negative words is also consistent with other research documenting that on average, linguistic tone is optimistic, on net (Frankel et al. 2010). Overall, there is substantial variation in firm characteristics measures, consistent with the sample representing a broad cross-section of firms.
Panel C presents industry composition of the sample firms, suggesting a relatively greater number of firms from the computer industry. The industry concentrations documented here are consistent with prior studies on earnings conference calls (e.g.: Mayew 2008).
4. CEO Attribution Measure I construct my attribution measure by counting how frequently a CEO uses firstperson and third-person pronouns. I assume that a CEO’s attribution of firm performance to internal (external) factors is characterized by more frequent use of firstthird-) person pronouns. Psychology theory suggests that personal pronouns reflect the attention allocation of the speaker (Davis and Brock 1975; Gunsch et al. 2000; Tausczik and Pennebaker 2010). For example, Gunsch et al. (2000) study the personal pronouns used in political ads and find that there are more first-person pronouns in positive ads and more third-person pronouns in negative ads, consistent with the intuition that positives ads focus on political candidates themselves and negative ads focus on their opponents.
In business contexts, how attributions are captured by personal pronouns is not as straightforward. Although I expect that a CEO’s explanation of firm performance will be characterized by the personal pronouns he or she uses, it is likely that personal pronouns may not completely or sufficiently capture internal and external attributions.
For example, Baginski et al. (2004) find that managers often mention general economic and environmental factors when making external attributions. Sometimes attributions are also characterized by words such as proper nouns (for instance, a CEO may attribute firm performance to the macroeconomic environment or specific companies.) Therefore, I conduct a construct validity test in section 5 to examine whether personal pronouns provide a satisfactory classification of internal and external attributions in business contexts.
I use LIWC 2007 to analyze the transcripts of earnings conference calls1. LIWC 2007 measures the percentage of words in categories defined by dictionaries in the software. This tool is widely used in the linguistics and psychology literature. I first isolate the words spoken by CEOs during the presentation and Q&A sessions of earnings conference calls using a Perl program. Then I obtain the percentage of personal pronouns from LIWC. The internal reference measure, INT, is the percentage of firstperson pronouns (including both singular and plural forms)2. Similarly, the external reference measure, EXT, is the percentage of third-person pronouns. Panel A of Table 2 lists the words included in the pronoun dictionaries in LIWC 2007.
My measure of management attribution, IvsE, is constructed as:
I use the natural logarithm to account for skewness and add 1 to both nominator and denominator to avoid division by 0.
LIWC 2007 is a text analysis tool developed by psychologists at the University of Texas at Austin and the University of Auckland. Li (2008) uses an earlier version of this software in his analysis of corporate annual reports. The software manual and more details can be found at www.liwc.net.
2 LIWC 2007 specifies both the singular and plural forms of personal pronouns. INT is the sum of I and WE;
The LIWC dictionary contains many derivatives of personal pronouns. For example, “Id” and “I’d” are included in the singular first-person pronoun list to account for the different spelling habits. Similar examples also include “we’ve”, “they’ll” and “they’ve”.
Table 2 (continued) Panel C: Linguistics Variables during Presentation and Q&A Sections
This table reports descriptive statistics of the linguistics variables of CEO speech for a sample of 50,434 quarterly earnings conference calls observations occurring between 2002 and
2007. Panel A lists the personal pronouns in the LIWC dictionary. Panel B presents the descriptive statistics of the linguistic variables; the Talking column shows the average lexical properties of people’s daily conversation from Pennebaker et al. (2007). Panel C presents the descriptive statistics of linguistics variables during the presentation and the Q&A section, respectively. Panel D reports mean values of these linguistics variables in each calendar year.
Panel E presents the mean values of IvsE in each LENGTH quintile. Panel F shows the correlations among the linguistic variables. ***/**/* indicates significance at 0.01, 0.05 and 0.10 level, respectively. Variables are described in Appendix A and are winsorized at the 1st and 99th percentiles.
Panel B of Table 2 presents the linguistic profiles of CEO spoken transcripts during earnings conference calls. To facilitate comparison with linguistic profiles of nonbusiness conversations, I provide in the last column of Panel B labeled “Talking”, the average lexical properties of people’s daily conversation documented in prior research (Pennebaker et al. 2007). Panel B shows that, on average, a CEO speaks 3,050 words during an earnings conference call. In a recent large sample study of earnings conference calls, Li et al. (2009) report that the typical length of a conference call is 32,484 letters and a typical CEO speaks 48% of them. Assuming that a typical English word contains 4.5 letters (Pierce 1980), the results in Li et al. (2009) translate to 3,465 words spoken by a CEO during an earnings conference call, which is comparable to what I find for my sample.
Panel B also suggests that business conversations exhibit several distinguishing patterns. First, CEOs use fewer personal pronouns in conference calls than people use in daily conversations (8.16 vs. 13.63)4, suggesting that business conversations are less personal than daily conversations. Psychology research suggests that the usage of personal pronouns reveals attention allocation of speakers (Tausczik and Pennebaker 2010). Conference call participants focus mainly on business events, which are less personal as compared to daily conversation. Second, CEOs are less likely to use singular first-person pronouns (I) in conference calls (1.24 vs. 6.30) but are more likely to use plural first-person pronouns (WE) in conference calls (5.42 vs. 1.09). This is reasonable because managers are talking in an organizational context with actions representing the inputs of many individuals. For example, a CEO is less likely to say “I decided to Lexical variables from LIWC 2007 are in percentage form.
expand in Asia” but rather to describe it in a group context such as “We decided to expand in Asia.” Third, CEOs use relatively fewer third-person pronouns than people do in daily conversations (0.56 vs. 2.30).
Panel C shows descriptive statistics of linguistics variables for the presentation and Q&A separately. Prior studies suggest that pronoun usage varies in different contexts. Pennebaker et al. (2007) analyze 168 million words used in different types of writing and speaking and find that the average number of personal pronouns used in “control writing” is 10.78% and the average personal pronouns used in talking is 13.63%.5 The lexical properties of managerial speech differ between the presentation and the Q&A sections. During a conference call the CEO usually reads a prepared statement regarding firm performance. Hence, it is reasonable that the transcript of a presentation is characterized by lower use of personal pronouns than that of a Q&A section where a CEO responds spontaneously to analyst questions. Consistent with this prediction, I find that the CEOs use more personal pronouns in the Q&A than the presentation (8.85 vs.
6.91). This is consistent with Pennebaker et al. (2007), who find that fewer personal pronouns are used in “control writing” than during conversations.
Panel D of Table 2 presents the mean values of the linguistics variables across my sample period. Although most linguistic variables show no time trends, conference calls Control writing refers to writing about non-emotional topics such as descriptions of ordinary events.
Talking refers to transcripts collected from individual talking in real world settings. These are analogous to presentation and Q&A sections of earnings conference calls.
have become longer over time. This is likely due to firms providing more concurrent and expanded disclosures surrounding earnings announcements (Francis et al. 2002), thereby requiring CEOs to discuss more during earnings conference calls. CEOs also tend to use more plural first-person pronouns over time (from 5.32 in 2002 to 5.54 in 2007), leading to an increasing trend of IvsE. Because of the time-series differences in linguistic variables, I control for time effects in my regression analysis by adding calendar quarter indicator variables.
Panel E shows the means and standard deviations (in parentheses) of the main attribution variable, IvsE, for different quintiles of LENGTH. The results suggest that different length of conference calls may affect the means and standard deviations of the attribution measure. IvsE decreases with the length of a conference call increases from quintile 2 to quintile 5. IvsE in the first quintile have much larger standard deviations than in other quintiles, because variation in personal-pronouns is amplified when the length of a conference call is short. Therefore, I include LENGTH as a control variable in my regression analysis.
The Pearson correlation matrix of the linguistic variables is presented in Panel F.
INT and EXT are negatively correlated with each other, reflecting the mechanical substitution effect between the internal and external attributions. The correlation between INT and LENGTH is negative, while the correlation between EXT and LENGTH is positive. This perhaps suggests that as conference calls get longer, a CEO speaks more of issues external to the firm than internal to the firm.
5. Construct Validity of the Attribution Measure Before proceeding to utilize the IvsE variable for hypothesis testing, I first attempt to provide construct validity. Providing construct validity is important for several reasons. First, IvsE is not a direct measure of internal and external attributions but assumes that internal attributions are characterized by more frequent usage of firstperson pronouns, and external attributions are characterized by third-person pronouns.
Second, it is possible that a measure based on pronouns is too noisy because internal and external attributions in business contexts may not be sufficiently captured by the use of personal pronouns. First-person pronouns may not include firm references and thus may not capture an internal attribution when a CEO refers his or her company as “the company” rather than using first-person pronouns. Third-person pronouns likely exclude external factors such as economic or environmental issues. If a manager refers to macroeconomic factors (e.g.: strength of the currencies, interest rates, and environment), these external attributions may not be captured by third-person pronouns. While these examples suggest the presence of measurement error, the degree of measurement error is not clear. Further, although theory and empirical evidence supports the use of personal pronouns to proxy for internal and external attributions, there is little prior evidence to support the use of personal pronouns to study attributions in business contexts.
Another reason to conduct a construct validity test is that personal pronouns may be correlated with some variables that are omitted in this paper. For example, Chatterjee and Hambrick (2007) use first-person singular pronouns to proxy for CEO narcissism. A narcissistic CEO may overestimate his or her role in the organization.