Submitted as part of the requrements for an SB in Cognitive Science, May 1987. Gently overseen by Muriel Cooper, of MIT's Media Laboratory, with occasional input by Susan Carey and Whitman Richards, of MIT's Cognitive Science Department. (Very slightly edited).
One Aspect of CRT Displays That Doesn't Make a Difference Studies have shown that people read from computer displays more slowly than they read from paper. While several theories have been offered to explain why, none has stood up to testing. One currently popular technique for enhancing CRT "readability" is known as "anti- aliasing." In this experiment I recreate the problem that anti-aliasing is supposed to correct in hardcopy, so its negative effects can be measured without the other inherent difficulties found in computer displays. Introduction As computers invade the work place, they bring with them Video Display Terminals - VDTs. While display technology is proceeding apace, at present the "D" in VDT usually means a Cathode Ray Tube - CRT. In the popular press CRTs have been linked to everything from eyestrain to miscarriage, however a serious body of inquiry into the cognitive effects of the devices did not exist until recently. A 1984 study (Gould and Grischkowsky, 1984) seemed to indicate most of the complaints about CRTs are unprovable, however during the course of the study one effect of the devices was strongly shown: People read 20-30% more slowly from CRT displays than from paper. Several subsequent studies have failed to pinpoint the reasons for the difference. My experiment likewise failed to find a sole cause for the CRT gap, but did show that at least one pet theory is probably false. A similar experiment (Gould et al 2/17/86) showed somewhat more ambiguous results. Past Studies - what the answer isn't CRTs differ from paper in many respects - Each of the following summarized experiments (all published by Gould et al 2/17/86) considered one or more aspects of that difference. 1 - Fixed orientation - 12 subjects in a proofreading task were shown to read equally fast (within margins of error) from paper fixed horizontally and paper fixed vertically. Both were read faster than CRT by 11 of the 12. [note - an earlier experiment (Tinker 1963 as quoted by Gould) showed a significant advantage of paper at a 45 degree angle to horizontal vs. paper at horizontal] 2 - Aspect Ratio - 18 subjects were asked to either proofread or read for comprehension (9 of each) three articles. Articles were presented either on CRT, on paper in normal orientation, or on paper rotated so as to be read "the long way." Subjects in the comprehension task were found to read paper "the long way" more slowly than normal paper, however in both tasks paper of either orientation was read significantly faster than CRT. The difference between paper normal and paper rotated was statistically significant in the comprehension task, however, so aspect ratio may be a co-factor in inferior CRT reading speed. 3 - Experience - Experiment 1 above was done with 6 novice and six experienced computer users. Experience subjects did have a slightly smaller CRT gap, but the gap still existed. The subjects in experiment 2 were all experienced computer operators and still showed a significant speed deficit between paper and CRT. 4 - Character spacing - Because character spacing on a CRT is greater than on paper, the eye must move a longer distance in order to traverse the same number of letters per fixation. [Eye fixations are made on a set number of characters, not a set visual angle - (Morrison and Rayner 1981, as quoted by Gould)]. This could increase reading time. However, research on office colleagues via "surprise" measurement by Gould et al showed people sit further away from CRTs than from paper, which largely compensates for the greater character spacing. (A more rigorous experiment by the same researchers did find more eye fixations per equivalent line of text on CRT vs. paper). 5 - "Flicker" - 9 subjects were asked to proofread 3 10-page articles, from either CRT, paper, or photographs of a CRT screen. Photos were read significantly slower than paper, though slightly faster than actual CRT. 6 - Polarity (white on black vs. black on white) - Ten experienced users proofread five different 5-page articles, in each of five different display formats - CRT (green on black), and black and white photographs of the CRT screen in both positive and negative and in two sizes. Reading rates from the CRT were found to be slightly slower than from the photographs, but the difference was not significant. No difference was found between photos of different polarities. 7 - Font - Sixteen subjects proofread 6 pages on paper and 6 pages on a Tektronix storage tube which displayed the identical font as that used on the paper (but which otherwise emulated a CRT). Layout, line width, and average words per page were identical between the two display mediums. Polarity however was inverse for the tube. Subjects read significantly faster from the paper. Past Studies - Combinations that help While it would appear that very little can be done to identify the relevant difference between paper and CRT much less rectify the situation, Gould et al (8/21/86) have found that when text is matched for font, polarity, size, color and layout, reading speeds on CRT approach that of paper. In addition to the above variables, Gould experimented with the use of anti-aliased character sets on CRT. His results were inconclusive. Anti-aliasing (and Purpose) Typically, CRT displays are made up of raster lines or dots (pixels). The use of this somewhat too large basic unit of display produces "jaggies" - rather than smooth outlines, characters on CRT displays have a "staircase" profile. The effect is called "aliasing." Aliasing has been considered as one of the causes for the CRT gap. Recently, a technique called "anti-aliasing" has come much into vogue among computer graphic artists. "Anti-aliasing" eliminates the jaggies by adding grey scale pixels to the outlines of the displayed character. The edges of the character appear to fade gradually into the background, and the human visual system provides a straight edge to compensate. While the resulting text is aesthetically more appealing, no hard evidence exists to show it has any effect on visual performance. My experiment replicates the "aliasing" effect on paper, so it may be compared to normal text (which will stand in for anti-aliased) without the additional handicaps of a CRT. Methodology Four pieces of text from the verbal section of the Graduate Record Exams (GREs), each 37 lines in length, were used. Each piece of text was typeset in Triumvirate (a san-serif font similar to that found on computer displays) 12 point with 14 point leading, on a 7 inch line, ragged right (so as to avoid spacing changes between letters). The copy was printed on high density photopaper on a Compugraphic 8600, at 1500 dots per inch. That typeset copy was then re-processed on an electronic scanner normally used for producing halftones (a system by which photographs are translated into dots large enough to be printed on an offset press), and re- printed on the 8600. The result was a digitally oversampled (ie "aliased") copy of the originally typeset material. The page layout, contrast and luminance, line widths, type size, and letter outlines were all identical to the original copy, however the scanned copy was apparently at a resolution of 50 dpi (the resolution of a poor quality CRT) not 1500. Booklets of alternating types of copy (reproduced on a high density xerox) along with xeroxed copies of the GRE questions were made up. Two pieces of text were designated for practice, and presented first in the booklet. All eight permutations of straight and jaggy, first or second, were represented. (see appendix). Eighteen subjects, 14 male and 4 female, in their late teens to late twenties participated in the experiment in return for a Ghirardelli Chocolate Square and the continued friendship of the experimenter. All had several years of previous experience using CRT displays. All had vision at or near 20/20 when corrected, though glasses wearers were given the option of reading with or without glasses, whichever was considered more comfortable. Subjects were seated in an unused conference room in the MIT Media Laboratory across the table from the experimenter, and asked to take a comfortable position. They then were instructed to choose a booklet at random from the collated pile, informed that they were going to take a section of the GREs, and asked to read for comprehension but as quickly as possible, and not to turn back to previously read material. Subjects were allowed to start at will. The experimenter used a digital watch equipped with a chronometer and lap-counter to measure the time between the turning of pages. Two tests were discarded, one due to mis-collation and one to an interruption during testing. Results No appreciable difference in time or accuracy was discovered for aliased vs. smooth text. Several subjects offered opinions as to the lack of "readability" of the aliased text, however. Discussion While more rigorous testing conditions may have uncovered a difference (it is clear from the range of the data that factors other than base reading speed had come into play), it is doubtful that anti-aliasing is the solution to the CRT problem. Similar experiments with other forms of degraded paper text (Gould, of course, 2/17/86) have also shown that degradation of paper fonts do not appreciably slow reading speed, though subjective reaction by subjects is usually negative. An experiment which used a visual performance task that did not require the reading of text (Booth et al) indicates the CRT gap may be due to overall resolution. The experiment was performed alongside a mental-rotation experiment, in order to determine if an anti-aliased display would affect performance on that task. Subject were asked to discriminate between 2 three-dimensional objects displayed in perspective on screen. The objects were made up of either 10 cubes or 9 cubes, and presented in a variety of orientations. While anti-aliasing improved speed in the task for low and medium resolution displays (below 171*171 pixels), at higher resolutions no anti-aliasing effects were found. Another as yet un-written-up experiment by Gould et al shows that high quality high resolution fonts on CRT read almost as fast as the same size and font on paper regardless of anti-aliasing (210 words per minute vs. 206 and 203 for anti-aliased and aliased respectively). In this latest experiment the text on paper and screen were made as identical in size and font as possible, and all had the same polarity. It is likely that as CRTs increase in resolution and therefore in their ability to display high quality fonts, their readability will increase regardless of aliasing. BIBLIOGRAPHY Booth, K.S., Bryden, M.P., Cowan, W.B., Morgan, M.F. and Plante, B.L.(1987). On the Parameters of Human Visual Performance/An Investigation of the Benefits of Antialiasing. Proceedings, CHI+GI '87 pp.13-19. Gould, John D., and Grischkowsky, Nancy (1982). Doing the Same Work with Hard Copy and with Cathode-Ray (CRT) Computer Terminals. Human Factors, 26(3), 323-337 Gould, John D., Alfaro, L., Barnes, V., Finn, R., Grischkowsky, N., Minuto, A. (2/17/86). Reading is Slower From CRT Displays than from Paper: Some Experiments that Fail to Explain Why. IBM Research Report as yet unpublished. Gould, John D., Alfaro, L., Finn, R., Haupt, B., Minuto, A. (8/21/86). Reading from CRT Displays Can Be as Fast as Reading from Paper. IBM Research Report as yet unpublished. Gould et al. Effect of Anti-Aliasing Itself on Proofreading. Preliminary write-up.