Investigating presentational change in U.K. annual reports: a longitudinal perspective.

Table 6 provides comparative data for the type of graphs used to portray key financial variables. In the 1989 sample, there was some diversity in graph type used. Although more than 80% used a column or bar graph, a sizeable minority used line graphs (4.5%) or other, more original, customized graphs such as pictograms (11.2%). However, by 2004, "normalization" of graph type had emerged. The standard presentational form had become the column or bar graph, representing 97% of all key financial variable graphs.

Table 7 reports the results of tests for graph selectivity in each sample year. Graph selectivity is said to occur when the use of a particular graph is contingent on "good" performance. Performance was classified as good or bad based on two alternative measures: (a) the direction of change in earnings per share in the current year, and (b) the direction of change in the particular variable graphed in the current year. Using each of these measures, selectivity was assessed in relation to the inclusion of at least one key financial variable graph, a sales graph, an income before tax graph, an earnings per share graph, and a dividend per share graph. The results indicate that, in general, selectivity continues to occur. This was particularly true for income before tax and dividend per share. Graph usage in these two variables was significantly associated with the change in earnings per share at the 5% level and with the change in the key financial variable itself at the 1% level. In both 1989 and 2004, therefore, graphs (with the exception of sales graphs in 2004) were more likely to be included when favorable, rather than unfavorable, performance was reported. Nonetheless, overall the results were less strong in 2004, indicating that graphs are being included on a less selective basis than in 1989. This perhaps reflects less concern with the key financial variable graphs generally as fewer were included in the 2004 annual reports.

Table 8 shows the length of time series presented in key financial variable graphs in 2004 compared to 1989. The majority of graphs in both years showed 5-year time series. However, the percentage showing this "norm" has declined from 72% in 1989 to 63% in 2004. In 2004, the more common alternative to this norm is a period less than 5 years. The pattern across all four key financial variables was similar. The percentage showing less than 5 years increased from 13% in 1989 to 27% in 2004; for sales and income there was an increase from 11% to 29% and 12% to 30%, respectively.

Interestingly, for this particular graph attribute the practice in 2004 has become more diverse than in 1989. A possible reason for this attribute is that in 2004 the economic cycle was at a stage in which shorter time series displayed more favorable trends. In 1989, the U.K. economic cycle reached a peak after a sustained period of growth lasting more than 5 years, whereas in 2004 there had been a small growth alter a 3-year decline (HM Treasury, 2005). In 1989, 5-year time series would show sustained increases. However, in 2004 cutting the time series to 3 years would offer a lower benchmark for comparison of current period performance. Thus, the incentives for management to impression manage in some cases appear to have overridden the desire to comply with reporting norms.

We investigate this interesting phenomenon in Table 9. In particular, we examine the relationship between company performance over the "normalized" period of 5 years and the management decision to include graphs and the number of years chosen. Specifically, distinguishing between companies whose performance improved and those whose performance declined, Table 9 reports the number of companies that chose to exclude graphs or present their key financial variables for a period of less than 5 years compared to those that presented the key financial variables for a period of 5 years. (12) The performance data were collected from the historical tables published in the annual reports (or from Datastream, when the former were unavailable). Consistent with Table 7, change in performance was measured in terms of a change in the performance of the earnings per share (Panel A) and in terms of a change in the performance of the key financial variable (Panel B) over the normalized 5-year period. The results presented relate only to the 2004 period because Beattie and Jones (1992b) did not undertake a similar evaluation.

Results in Panels A and B indicate management tended to present information in a positive light. Specifically, companies whose performance had declined over the 5-year normalized period were more likely to either omit the key financial variable (selectivity) or present data for a period of less than 5 years as compared to those companies with improved performance. Results for pretax income and dividend per share were striking, as were those for all key financial variables combined. Results for earnings per share were also statistically significant, although those for sales were not. The sales result perhaps confirms the earlier evidence that this key financial variable is declining in importance. Moreover, results of a further analysis restricted to graph users only (see note to Table 9) confirm statistically significant selectivity in the number of years graphed at an aggregate level: Graph users with favorable performance were more likely to chart graphs for 5 years than graph users with unfavorable performance. Unfortunately, the chi-square values for this analysis were not valid for all of the individual key financial variables when the sample was restricted to only graph users because of the resulting smaller sample sizes. Collectively, these results explain why the 5-year period over time has not become normalized and also identify a new way in which companies engage in impression management.

Tables 10 to 12 report on aspects of graph measurement distortion in key financial variable graphs. Table 10 shows the incidence of materially discrepant graphs, using a materiality threshold of both 5% (following Beattie & Jones, 1992b) and 10% (the level at which Beattie and Jones, 2002, found user perceptions to be distorted). At the 5% cutoff (Panel A), 30% of graphs exhibited material distortion in 1989, rising to 60% in 2004. At the 10% cutoff (Panel B), the corresponding figures were 20% in 1989 and 49% in 2004. Clearly, therefore, the incidence of material distortion has risen dramatically. This increase may be associated with the decline in selectivity noted above. If companies feel less able to avoid including graphs showing unfavorable performance trends, they may distort the graphs to reduce the unfavorable impression conveyed.

Several further observations can be made from Table 10. In both years, all the key financial variable graphs, except for income, were more likely to be materially exaggerated than materially understated, although the relative incidence of material understatement had risen markedly by 2004. In both sample years, income before taxes was the one key financial variable not to be exaggerated. (13) By 2004, users' perceptions of key financial variable performance were likely to be affected in approximately half of the cases (Beattie & Jones, 2002).

Table 11 gives a breakdown of the magnitude of measurement distortion found in key financial variable graphs. It is apparent that the greater incidence of material measurement distortion reported in Table 10 above seems to occur especially at the extremes. In the 2004 sample, 12% of key financial variable graphs contained distortion in excess of 100%, compared to 3% in the 1989 sample. At the other extreme, 6% of key financial variable graphs contained distortion below -50%, compared to 1% in the 1989 sample. Moreover, if the number of GDIs exceeding 25% is taken, the difference between the two samples is further emphasized. In 1989, only 11% of the sample had GDIs greater than 25%; however, by 2004 it was 35%. Finally, Table 12 indicates the cause of material distortions. It appears that, by 2004, the obvious, identifiable causes of distortion (e.g., the use of a nonzero or broken vertical axis or a non-arithmetic scale) have disappeared, leaving behind more subtle distortions. For example, graphs with identifiable non-arithmetic scales have now been replaced with graphs with no stated scales--distortion here is easier to conceal. In addition, in a small number of cases, companies have failed to indicate the precise values that are being graphed on their columns or bars. With relatively small graphs being used to chart what can be widely varying values, there is thus heightened scope for misinterpretation.

SUMMARY AND CONCLUSIONS

There is a paucity of research that examines the manner in which the discretionary elements of corporate annual reports have evolved over time. These discretionary elements concern the overall structure and form of the annual report and the usage of narratives, graphs, and pictures. The present study contributes to the limited literature in this area. There is a particular focus on graphs, which represent an important presentational format. New evidence is presented by analyzing additional aspects of an existing sample of corporate annual reports from 1989 (Beattie & Jones, 1992a, 1992b) and an up-to-date, fully comparable sample from 2004. Longitudinal comparisons are made with the findings of Beattie and Jones (1992b) and Lee (1994). In the latter case, comparisons were made using a restricted "large company" sample and the full sample of 1989 and 2004 reports.

The main findings relating to structure and format over three decades were as follows when compared with the full sample.