The effect of knowledge on accuracy and partisanship on distortion of memory of baseball statistics


In order to study the effects of expertise, time delay and partisanship on memory distortion, 34 college students watched a baseball game and then recalled statistics from the game directly after and again, one week later.  Two tests were performed.  In the first, subjects were grouped according to their reported expertise and experts were found to be more accurate than novices at recalling statistics.  In the second, subjects who reported partisanship toward one team did not consistently distort toward the their team.  Instead, distortions were mostly toward average numbers for individual statistics.  Distortion in this study seemed to be due to filling in of more generic ideas rather than emotional gratification.


Do people tend to distort hard-to recall information in their favor?  The literature so far says yes.  In Bahrick, Hall and Berger’s (1996) study, college students tended to distort memories of their high school grades upward.  This finding is attributed in part to more frequent rehearsals of positive content, but because students who got mostly A’s were much more likely to distort a forgotten grade to an A, some of this correlation may be due to reconstruction based on generic memories.  Their work dealt with the differences between quality-oriented and accuracy-oriented studies of memory, as suggested by Koriat and Goldsmith (1994).  This investigation follows their model of study to some degree.

Another related question must be asked: what is the affect of expertise and time on memory distortion?  Some research (Sanbonmatsu, Sansone & Kardes, 1991) has suggested that only moderate inferences are drawn shortly after initial processing of the information, and that stronger influences were made after an extended period of time.  Expertise also affected inferences, with people more knowledgeable in a subject area less likely to draw inferences about unknowns than novices.  The study, however, dealt with drawing conclusions based on a lack of knowledge rather than attempts at recall.

Recall of a baseball game based upon common statistics is a suitable area for exploring these questions.  Within just three innings, enough data can be collected to produce worthwhile results.  Subjects watching the same game in the same room will have very similar encoding conditions and statistics that have definite positive and negative directions for each team are easily identified.  Extensive statistics are kept on all baseball games, so accuracy is easily verified.  More importantly, a wide range of fans (who tie great value to their teams performance) and non-fans (who have less reason for distortion) for each team are easy to find.  It is believed that the emotional effect of the game will be magnified if a World Series game is used.


Thirty-four Ohio Wesleyan students participated in the experiment to fulfill a requirement in their Introduction to Psychology class.  There were 15 male and 19 female participants.  The possibility for participation was open to any enrolled student.

The participants watched the first three innings of game 4 of the 1997 World Series.  They were allowed to converse and interact about the game naturally, but no one was allowed to leave the room during the game.  At the end of the third inning, they were administered a questionnaire about specific statistics from those three innings, their level of partisanship towards the teams and their baseball expertise.  Participants were asked to rate themselves on a scale of 1 to 7 on their partisanship, with 1 indicating avid Indians fans, four neutral, and 7 indicating avid Mariners fan. Participants were asked to rate their own baseball expertise based on a scale of 1 to 5, 1 indicating little or no knowledge, 3 average and 5 expert.

The questions on team statistics were designed to be difficult to directly calculate or remember but readily estimated.  The questions were on statistics important to each team’s performance.  Statistics for each team included number of stolen bases, number of runners left on base, total number of pitches to that team, total number of strikes pitched to the team, total number of balls pitched to that team, number of assists for infielders and number of put outs for outfielders for each of the three innings.  Participants were also asked to approximate each team’s batting average over the three-inning period.  Participants were asked to fill out the form to the best of their ability and were not able to converse or compare answers.  One week later students were called back and asked to complete a form with the same statistical questions.  Again, they were not allowed to compare answers during the test.  A copy of the questionnaire is Appendix I.

Actual statistics from the game were verified at

Results and Discussion

Knowledge versus accuracy

In order to study the relationship between knowledge and accuracy the absolute difference between reported statistics and actual statistics was computed in each statistic for every participant.  Subjects who reported baseball knowledge of 4 or 5 were assigned to the expert group and those with knowledge of 1 or 2 were assigned to the novice group and their absolute differences for each statistic were then averaged for each session (figure 1a).  Putting participants into expert and novice groups provided a larger subject base for later testing.  The statistics were tested with 2 x 2 ANOVAs, with session as a within subject factor and baseball knowledge as a between subject factor (Table 1).

Session 1 Session 2
D.V. Novice Expert Novice Expert Session F Know F Interact F
ia1 1.4167 0.8462 1.4545 0.7692 0.11 *5.58 0
ia2 1.9167 0.9231 3 0.6154 0.34 ***14.24 1.41
ia3 2.5 0.7692 2.4545 1 0.1 *7.19 0.54
ilob1 0.4286 0.3846 0.6923 1 **11.22 0.82 1.25
ilob2 0.6429 0.5385 0.5385 0.6154 0 0 0.19
ilob3 0.6429 0.3077 1.4615 1.3077 **11.54 0.74 0.2
ipo1 2 0.8462 2 1.3077 0 3.18 0.75
ipo2 2.0833 0.5385 2.7 1.1538 1.62 **11.72 0.07
ipo3 1.8333 0.3846 3 1.3846 3.77 **8.12 0.01
isb1 0.2143 0.0769 1.1538 0.3077 *6.25 3.38 2.25
isb2 0.5 0.1538 1.2308 0.7692 **11.19 1.89 0.04
isb3 0.7857 0.1538 0.6667 0.4804 0.06 4.09 0.98
ma1 1.75 0.5385 2.0909 0.6154 0.37 ***18.22 0.12
ma2 1.5 0.6923 2.3636 0.6154 0.4 *6.66 0.61
ma3 2.9167 1.1538 2.2727 1.5385 0.41 *6.02 *4.35
mlob1 0.9286 0.5385 0.5385 0.5385 1 0.57 1
mlob2 0.5 0.3846 1.0769 0.7692 **7.74 1.01 0.41
mlob3 0.6429 0.5385 1 0.6154 1.17 1.48 0.42
mpo1 2.25 0.8462 2 1.2308 0.02 3.92 0.84
mpo2 1.25 0.5385 1.7273 1 1.53 2.61 0
mpo3 1.8333 0.6154 2.2727 1.5385 2.06 4.27 0.61
msb1 0.2857 0.0769 1.6154 0.0769 *6.24 *7.12 *6.24
msb2 1.0714 0.3077 1.3846 0.6923 2.35 3.36 0.03
msb3 1 0.0923 1.3846 0.8462 1.47 1.85 0.27

This test showed that more expert participants were more accurate and that accuracy decreased over time.  As expected, the expert subjects reported numbers with a lower average deviation from the actual statistics than the novice subjects in every significant case (Table 1).  Experts were generally more accurate for the other statistics as well.  The effect of time on accuracy was also as expected; for both groups the subjects’ data was less accurate in the second session than the first in nearly every significant case (Table 1).  There seemed to be no pattern of significant interaction between the two independent variables.  It is concluded then that experts were more accurate than novices and that experts had no advantage over novices in losing accuracy over time.

Partisanship versus memory distortion

The distortion of each individual was defined as a difference between the reported and actual statistic in a direction that would have benefited one of the teams.  For each team, an inflation of the number of assists, balls pitched to the team, and number of pitches to the team was considered positive for that team and a deflation of strikes pitched to that team was considered positive for the other team.  No other statistics were included in this study in order to avoid floor and ceiling effects.  The number of participants who distorted toward each team was recorded for each statistic in each session.  In order to show the relation between reported partisanship and distortion, these data were tested twice: first for all participants reporting any partisanship toward the Indians (partisanship scores of 1 through 3)(Table 2) and then for all participants reporting extreme partisanship toward the Indians (partisanship score of1)(Table 2B).  The numbers of subjects distorting toward each team were tested with a chi-square to determine statistical significance, assuming that if no systematic distortion existed, half would distort toward each team.

Session 1 Session 2 Statistics
Dependent Variable For Ind. For Mar. chi-square For Ind. For Mar. chi-square actual stat avg stat
Indians infield assists, 1st inning 7 15 2.9091 11 10 0.0476 2 1.666667
I. infield assists, 2nd inning 6 12 2 12 7 1.3158 2 1.666667
I. infield assists, 3rd inning 17 5 *6.5455 20 2 ***14.7273 1 1.666667
I., balls pitched to, 1st inning 11 17 1.2857 15 12 0.3333 11 13
I., balls pitched to, 2nd inning 24 6 ***10.8000 22 4 ***12.4615 7 13
I., balls pitched to, 3rd inning 8 23 **7.2581 2 27 ***21.5517 21 13
I., strikes pitched to, 1st inning 14 8 1.6364 11 14 0.36 10 7.333333
I., strikes pitched to, 2nd inning 0 31 0 29 2 7.333333
I., strikes pitched to, 3rd inning 10 17 1.8148 6 17 *5.2609 10 7.333333
I. runners left on base, 1st inning 5 5 0 5 15 *5.0000 1 1.666667
I. runners left on base, 2nd inning 7 12 1.3158 9 7 0.25 1 1.666667
I. runners left on base, 2nd inning 3 1.666667
I., number of pitches to, 1st inning 5 26 ***14.2258 10 17 1.8148 27 29.66667
I., number of pitches to, 2nd inning 24 5 ***12.4483 27 1 ***24.1429 14 29.66667
I., number of pitches to, 3rd inning 5 26 ***14.2258 4 24 ***14.2857 48 29.66667
Mariners infield assists, 1st inning 7 15 2.9091 2 18 ***12.8000 1 0.666667
M. infield assists, 2nd inning 4 14 *5.5556 2 17 ***11.8421 1 0.666667
M. infield assists, 3rd inning 0 0.666667
M., balls pitched to, 1st inning 9 21 *4.8000 6 20 **7.5385 6 9
M., balls pitched to, 2nd inning 13 17 0.5333 7 19 *5.5385 7 9
M., balls pitched to, 3rd inning 13 18 0.8065 8 20 *5.1429 14 9
M., strikes pitched to, 1st inning 0 3.666667
M., strikes pitched to, 2nd inning 27 3 ***19.2000 27 2 ***21.5517 4 3.666667
M., strikes pitched to, 3rd inning 24 6 ***10.8000 21 5 **9.8462 7 3.666667
M. runners left on base, 1st inning 8 7 0.0667 8 7 0.0667 1 1
M. runners left on base, 2nd inning 9 6 0.6 14 8 1.6364 1 1
M. runners left on base, 2nd inning 8 12 0.8 9 7 0.25 1 1
M., number of pitches to, 1st inning 3 27 ***19.2000 2 25 ***19.5926 11 19.33333
M., number of pitches to, 2nd inning 10 21 *3.9032 4 22 ***12.4615 15 19.33333
M., number of pitches to, 3rd inning 26 4 ***16.1333 20 7 *6.2593 32 19.33333

Despite the fact that all subjects in the first test group reported some amount of partisanship toward the Indians, the data does not show a systematic distortion in favor of the Indians.  In the first session, the group distorted toward the Indians in only six out of the thirteen significant statistics and in the second session only 6 out of the 17 significant statistics.  This does not support the theory that memory is distorted in the direction most pleasant for the subject.

In order to explain the pattern of distortion, the actual values for the statistics were examined (Table 3).  In 11 out of the 13 dependent variables that showed significant distortion in the first session and in 12 of the 17 in the second session, most subjects distorted toward the average of that statistic over the three innings.  If the average over the three innings can be taken to approximate a generic expectation for that statistic, this data supports the conclusion that most subjects distorted toward generic values for most statistics.  This tends to support the theory distortion is caused by filling in generic information where specific memories have been lost.

The second test group, made up of the 13 students who reported very strong partisanship towards the Indians, did not significantly differ from the first test group (Table 2).  Although there are too few cases of significance to draw solid conclusions, the data from the second group seems to conform to the findings of the first group.

Interaction between knowledge and partisanship

There were too few participants to determine the effects of interaction between distortion and partisanship (Table 4).

General Discussion

The main limitation of this study was the low number of participants.  More significant evidence could be gathered with a larger subject base and generic groupings of ranges of expertise and partisanship could be avoided in order to look for more subtle differences.  Also, the interaction between knowledge and partisanship could be better studied.

Another limitation was the fact that many of the subjects, especially novices, may not have thought about these particular statistics during the game and were therefore less directly exposed to the content than others. Also, although baseball fans are interested in their favorite team’s performance, watching a game is still more passive than participating and this may be reflected in the lack of evidence for systematic distortion in the direction of the favored team.  Involving subjects directly in a sport where statistical results reflect personal performance might have a better chance of showing this distortion.  A sport such as bowling would be good for this study, involving participants directly as well as exposing them directly to statistical measurements of performance via score sheets.  Enough statistics are collected during a bowling game that subjects will have a general impression without giving them exact memories of the numbers and these statistics are easily collected and verified.


Bahrick, Harry P., Hall, Lynda K., & Berger, Stephanie A.  (1996).  Accuracy and distortion in memory for high school grades.  Psychological Science, 7, 265-271.

Koriat, A., & Goldsmith, M.  (1994).  Memory in naturalistic and laboratory contexts: Distinguishing the accuracy-oriented and quality-oriented approaches to memory assessment.  Journal of experimental Psychology: General, 123, 297-315.

Sanbonmatsu, David M., Sansone, Carol, & Kardes, Frank R.  (1991).  Remembering less and inferring more: Effects of time of judgement on inferences about unknown attributes.  Journal of Personality and Social Psychology, 61, 546-554.