Content-Type: text/html
Legibility of Serif Type
for Use as Body Copy
in Computer Mediated Communication
By Professor Joel Geske
Iowa State University
126 Hamilton Hall
Ames IA 50011
[log in to unmask] y (515) 294-0477
Submitted for Consideration to
AEJMC Visual Communication Division
April 1996
Abstract
This experimental design used 107 subjects to test legibility of
Palatino type in three sizes and three faces viewed on a computer monitor.
The study found 12 point type most legible with 10 and 9 point
significantly less legible. Bold face in general enhanced legibility, although
results were mixed. In all cases bold was a better choice for emphasis than
italics. Italics should be used with extreme caution and avoided a smaller
sizes.
Legibility of Serif Type
for Use as Body Copy
in Computer Mediated Communication
Numerous studies have been done determining both legibility and
readability of type in printed materials in mass communication settings (Tinker,
1963, McVey & Weigeshaus, 1973 and Felici). However, a great many of the
classic studies in readability and legibility are from the 1920-1940 era.
Starting in the 1980's with the advent of the personal computer and in
particular in the mid-1990's with the increased usage of the Internet and other
computer mediated sources for transferring large amounts of information it is
necessary to see if many of the basic rules of legibility and readability of
type hold true for computer monitors.
Review of Literature
The terms legibility and readability have quite distinct meanings
although they are inter-related and sometimes used interchangeably. Prior to
1940 the term "legibility" was used to discuss factors of ease and speed of
reading. After 1940 some researchers began using the term "readability" as a
broader and perhaps more meaningful term with legibility referring to the
recognition of letter forms. However, with the advent of readability formulas
for measuring the difficulty of reading material there has been some confusion.
For the purposes of this study, the term legibility will be used to measure the
human eye's ability to discern characters and words--to actually be able to
correctly recognize the letter and word forms. Readability will be reserved for
measures testing the ease of reading and understanding material(Tinker, 1963).
Basic notions about text and readability have developed over many
centuries. One of the first recorded legibility tests was done in France in the
1790's. However, two major discoveries occurred around 1900 are among the most
important concerning legibility and readability:
1) Experienced (not beginning) readers read in whole words, not a
character at a time and,
2) Readers use saccadic leaps or jumps along a line of text and
pause
to read a regular intervals. Readers tend to read several words
or a phrase at one time before the eye moves to the next
grouping of word.
In a very fundamental way, these discoveries helped shift research
from legibility of single characters to readability.
Some legibility problems are inherent in the design of the alphabet
itself. In fact, some of the most used letters in the English alphabet are easy
to confuse, including C - G; H - N; E - F in the upper case and c - e (along
with a and o in some faces); b - d; and p-q in the lower case. On computer
screens i - l - I - 1 (small i, small l, capital i and numeral one) can also
cause confusion.
Many type and design books suggest rules for legibility. Serif type
faces are generally considered more legible than sans-serif type faces due to
the added information to the eye that the serif provides as well as the "line"
the serifs provide to guide the eye. However, studies cited by Tinker in 1963
on studies conducted by Tinker and Paterson showed no statistically significant
differences in readability of ten serif type faces. Two other studies in the
1960's tested sans vs. serif faces and showed no significant reading problems
for the sans serif types (Tinke, 1996). At best, the research suggests that
serif type faces may be more legible, but other studies indicate that properly
typeset and formatted that sans serif faces do not pose serious legibility
problems.
The computer screen monitor, however, brings additional problems to
the issue of legibility of type. When graphics based fonts first started
appearing in graphical interfaces such as Macintosh and later Windows
applications, they were almost always stored and displayed in bitmap format. A
typical computer monitor has 72 pixel resolution (or 72 pixels by 72 pixels in
each square inch of screen) for Macintosh and 96 pixel resolution for PC's. The
minimum stroke width is one pixel wide or high. The next possible stroke width
is double that or two pixels wide. Especially in smaller size fonts (such as
those used for body copy) this allows for very little leeway in displaying
fonts. Bitmapped fonts can be enlarged, but only by multiplying the existing
pixels which generally results in the "jaggies" or jagged edges on the enlarged
type. Likewise, type can be reduced, but only with the problem of open areas in
some letters filling in or some important pixels (such as the cross arm of an A)
being deleted. Therefore, most bitmap fonts are designed for a particular size
even though this limits type size choices.
To compound the problem, when text is highlighted and the bold
function is used to create bold face type, the computer generally doubles the
vertical stroke width of the letter but leaves the horizontal stroke of the
letter at the original height. This helps solve the problem of letters filling
by not adding additional pixels on the horizontal stroke, but alters the
character and design of the type on a fundamental level. Doubling the vertical
pixels also has the effect of extending the serifs and causing them to run
together. Even sans-serif types can have a problem with letters crowding or
running together in a bold face.
"Italic" type also has inherent problems on screen. In fact, rather
than true italic, most monitors actually display an oblique or slanted form of
the standard face. Again, especially in smaller font sizes that are using
letter strokes only one or two pixels wide, this slanting letter can take on a
very jagged and difficult to read appearance.
Serif type in particular may present unique problems. One type
designer (Chuck Bigelow, designer of the Lucinda Family or type) comments,
"when printed the serifs on typefaces are only a tiny percentage of the typeface
design. But on-screen, in order to display the serifs using the limited number
of available pixels, they take up a much bigger proportion of the information
than they do in print. Serifs should be small things--but on screen they become
big...noise or distracting chunks of interference (Petzgold). Especially in
bold and italic faces, serifs can run together, overlap and cause legibility
concerns.
More recently, outline font technology has come to the personal
computer. Here the letter is outlined using a mathematical formula to create
straight lines and curves. Once the outline is complete it is filled in solid.
One problem, however, is that this method is much slower to produce a page full
of text than bitmapped fonts. To solve this problem, once the font has been
formed, it is converted to a bitmap form and saved or "cached" in the computer
memory. This conversion is known as rasterization. Conversion takes some time,
but once completed the display fonts can be as fast as the display of bitmapped
fonts. Unfortunately, in the rasterization process some rounding off of numbers
occurs. Some letters, such as an H, which should have two vertical strokes of
about the same width may end up with one a pixel larger. On large display type
or when using a 300 dot per inch printer it may be hardly noticeable. On
screen, however, with only a 72 or 96 dot resolution and when using small type
faces the problem can be dramatic and severely affect both legibility and
readability (Petzold).
In addition, while outline and advanced technologies like Adobe Type
Manager (ATM), TrueType or Speedo use outlines and gray scales to smooth out
letters they also cause additional problems. These technologies work well with
larger type, but readers spend relatively little time on headlines and much more
on body copy. For body copy, the difficulty in reading comes from fuzzy edges
and making out characters. Gray scale and anti-aliased type increase the
"fuzziness" and the eye has to battle to draw a sharp focus. (Felici, 1996).
It can be concluded that type viewed on computer screen monitors
present unique challenges to the typographer and designer. Many of the
assumptions made for print legibility and readability may not hold true for
on-screen type. In particular, serif type faces may actually develop legibility
problems in bold and italic faces where the information normally added by
serifs may in fact become "noise" and hurt legibility.
Even ignoring the fact of a computer screen being luminous while
traditional type studies were done on reflective print media, the very nature of
type construction and display on the screen bring into question legibility
factors. Simple legibility factors must be considered before serious research
can be conducted on readability.
The Research Question
As more traditional media look to computer mediated delivery
systems, the question of how to provide large amounts of information in the most
legible, readable form becomes an important factor. Today, with the increasing
popularity of the Internet, we find books, magazines, newspapers, advertising
and even videos on the computer. The fundamental question is to look at type
legibility on screen to determine type styles and type sizes that are most
legible.
The review of literature and the exploration of problems encountered
with screen resolutions does not allow one to draw conclusions based on previous
research concerning serif faces. While previous research in print suggests
serif faces are easier to read, problems with screen resolution, especially when
making faces bold, would suggest that serif faces may have significant problems
when used on computer monitors. Serif type faces tend to use more thick/thin
differentiation in strokes that can lead to rasterization problems and increased
problems with bowls (or spaces in letters such as "e") filling in. Finally,
serifs can run together, especially on bold type. For these reasons, a serif
type face will be explored for a study on the effects of the variables of size,
boldness and italics on legibility of a type face.
This study looks at three basic research questions:
1) What is the most legible type size for body copy or long
segments
of text?
2) Does bold facing enhance or inhibit legibility at
different type
sizes?
3) Do italic faces affect legibility?
Methodology
Tinker provides a comprehensive discussion of methodologies for
testing legibility (Tinker, 1963). This study will use the Speed of Perception
method for testing character legibility. By using a very short exposure
technique, the quickness and accuracy in perception of words, symbols and
phrases can be measured. This method has been used successfully in previous
research to study factors such as use of serifs, boldness of letters and optimal
width stroke and optimal height/width ratios of letters (Tinker, 1963).
In this method, the human subject observes a pre-exposure field, is
briefly exposed to the message and ends with a post-exposure field. The
exposure to the message is very short, generally in the range of 1/10 of one
second to 1/100 of a second.
For this experiment, the subject is seated in front of a computer
screen. The screen has a black background with a white rectangular field in the
upper third of the screen at approximately eye level. Below the rectangular
field is a button for the subject to press when (s)he is ready for the exposure.
The computer is programmed to provide 1/60 of one second exposure before
moving to the post exposure field to block the words. This time of exposure was
pre-tested to find a time that provided differentiation between type styles and
sizes. This exposure time yields a single act of vision since the timing
prevents a second fixation or saccadic leap. Older studies used a tachistoscope
to view type printed on cards. This experiment uses the computer screen for the
subjects to view. For this test, Macintosh PowerPC 7100/80AV computers were
used with color monitors.
For this experiment the researchers used black type on a white
background. This is considered one of the most legible combinations in print
studies. While research suggests black on white may not be the most readable
for long text on computer screens, this experiment uses short phrases and short
exposures and each type face and style is being viewed with the same color
combination.
Palatino was selected as a sans serif type face as it is commonly
available on many computers. It has relatively long acenders and decenders and
a moderate serif size to provide good letter and word recognition. It also has
a fairly open design in letters such as "a, e and o" to minimize problems with
filling in of "bowls" or open areas of the type.
Three body type sizes, 9 point, 10 point and 12 point, were
selected. Sizes below 9 point were considered. However, when type set in sizes
smaller than 9 point are magnified, one can clearly see that many letters lose
all definition and simply become square blocks of pixels. Sizes over 12 point
were not considered efficient for body copy.
Each type size was tested in normal, bold and italic. The serif
faces were mixed with other type faces in a variety of sizes and styles.
This arrangement yielded 9 cells for Palatino in the trial. For each
exposure a phrase was inserted in the exposure field. Each phrase consisted of
three common words that form a simple sentence such as "Trees have leaves" or
"Dogs chase cats". Phrases were chosen as most adults take in several words in
a saccadic leap or fixation. Each phrase had 15, 16 or 17 (16 plus or minus
one) characters and spaces.
Phrases were randomly assigned to type faces and styles and each
type style was randomly selected for order of testing. To help control for
difficulty of reading a particular phrase, two forms were tested with a second
random assignment of phrases to type face. No phrase and type face combinations
were repeated from Form A to Form B.
Subjects were given several examples to adjust to the procedure and
methodology before the test began. Subjects controlled the rate of testing by
using the mouse to click on a button directly below the exposure area. Subjects
recorded each word in the phrase that they could perceive on a form that could
be checked for accuracy and coded.
Subjects chosen were students at a major midwestern university in
the United States. College students were chosen for several reasons. First, one
of the largest and fastest growing areas of computer mediated communication is
the Internet. One of the most comprehensive and current surveys on Internet use
still shows students making up over half of the total users (even with the
increased number of commercial users) ( Matrix News, 1995). College students in
general are literate, familiar with computers and are at an age where there are
minimal uncorrected vision problems. This should be viewed as an optimal
audience, however, with differences expected with less literate subjects or at
different age groups.
A total of 122 students were recruited for the experiment.
Students were volunteers from a Principles of Advertising class with over 10
majors represented in areas of business, communication, design, human
performance and consumer sciences among others. Results from 15 respondents
were not used because they had learning disabilities that interfered with
reading or English was not their primary language (since this was a speed of
recognition test, it was deemed that it was not appropriate to include these
results in the findings.) This left a sample of 107.
Mean age was 21.3 with: 3 subjects age 18; 12 subjects age 19; 36
subjects age 20; 35 subjects age 21; 8 subjects age 22; 5 subjects age 23; 2
subjects age 24; 4 subjects age 25; 1 subject age 26 and 1 subject age 32.
Gender for the respondents was 56% female and 44% male or just about flipped
from the educational users of the Internet which are about 59% male and 41%
female.
Students were assigned to Form A or Form B as they entered the
testing lab.
Findings for a Serif Type Face (Palatino)
The legibility tests for Palatino were run with two test groups.
One concern was that legibility could be affected by the phrase chosen rather
than type characteristics. Therefore, each type size and style was replicated
using a different randomly selected phrase. A significance level of p=0.01 was
chosen as a conservative measure to make sure the groups are not statistically
different due to phrase chosen. Of the nine cells, 8 showed no significant
difference at the 0.01 level, as shown in Table 1.
Table 1
n=54
n=53
Type Style
Sample A Mean
Sample B Mean
T-test
p value
Significant at 0.01 level
Palatino 12
2.35
2.43
p=0.630
no
Palatino 10
1.32
1.64
p=0.120
no
Palatino 9
1.48
1.11
p=0.090
no
Palatino 12 bold
1.82
2.23
p=0.031
no
Palatino 10 bold
2.28
2.06
p=0.300
no
Palatino 9 bold
1.33
2.26
p=0.000
yes
Palatino 12 italic
1.96
1.64
p=0.120
no
Palatino 10 italic
0.70
0.89
p=0.260
no
Palatino 9 italic
1.06
1.45
p=0.029
no
Overall, there is little significant difference in Sample A and
Sample B for most type faces. However, it does appear that the words or phrases
may make a difference in the smallest faces in bold type. For the remaining
tables, caution should be used in any pooled data comparisons for the 9 point
bold type face.
Given that Sample A and Sample B are not statistically different,
the question of differences in type style and size can be addressed. When
comparing populations, the researcher wanted a very conservative measure to
ensure the sample populations were equal and used a significance level of
p=0.01. Rather than pool data however, this study will first present the
information as two studies (Sample A, n=54) with a replication (Sample B,
n=53) to look for any unanticipated differences and to corroborate findings. If
the results between groups is not conclusive, and to add power to the t-tests,
data will be pooled to look at overall trends with the larger sample.
When comparing type the researcher was looking for broader trends
and differences between sizes and styles. Here a significance level of p=0.05
was used.
Comparison of Palatino Type by Size
Null Hypotheses (NH)
NH1: There is no significant difference between Palatino 12
point
and Palatino 10 point type viewed on computer
monitors.
NH2: There is no significant difference between Palatino 10
point
and Palatino 9 point type viewed on computer monitors.
NH3: There is no significant difference between Palatino 12
point
and Palatino 9 point type viewed on computer monitors.
As Tables 2 and 3 indicate, the null must be rejected in two cases,
NH1 and NH3. Palatino 12 point type is significantly more legible than both 10
point and 9 point type in both the first test and the replication. Palatino 10
point type was not found to be significantly more legible in Sample A, but was
in Sample B. Pooling the data here shows no significant difference (n=107;
Palatino 10 mean = 1.48; Palatino 9 mean = 1.30; p=0.24) and we accept the null.
Table 2 Sample A y Palatino Normal Type by Point Size
n=54
Type Style
Mean # words
T-test
p-value
Significant
p=0.05
Palatino 12
2.352
p=0.000
yes
Palatino 10
1.310
p=0.44
no
Palatino 9
1.480
p=0.000
yes
Palatino 12
2.352
Table 3 Sample B y Palatino Normal Type by Point Size
n=53
Type Style
Mean # words
T-test
p-value
Significant
p=0.05
Palatino 12
2.434
p=0.000
yes
Palatino 10
1.640
p=0.013
yes
Palatino 9
1.110
p=0.000
yes
Palatino 12
2.434
Comparison of Palatino Bold Type by Size
Null Hypotheses (NH)
NH4: There is no significant difference between Palatino
Bold 12
point and Palatino Bold 10 point type viewed on
computer monitors.
NH5: There is no significant difference between Palatino
Bold 10
point and Palatino Bold 9 point type viewed on
computer monitors.
NH6: There is no significant difference between Palatino
Bold 12
point and Palatino Bold 9 point type viewed on
computer monitors.
Results are mixed ( see Table 4 & 5). Recall that there were
significant differences found between Sample A and Sample B in the 9 point Bold
area.
In Sample A the null must be rejected in all three cases. There are
significant differences in the mean number of words between 12 point Bold and 10
point Bold with the 10 point Bold being more legible. There are also
significant differences in the 10 point/9 point and 12 point/9 point
comparisions with the larger type size being more legible in each of those cases
(Table 4).
For Sample B, the null can not be rejected for any of the
comparisons. There appears to be no significant differences (Table 5).
In this case, results will be pooled. A larger sample size will
help to generate a truer picture of the normal curve for each sample for t-test
comparisons. Caution does need to be used with the 9 point Bold samples.
With the pooled data (Table 6) Palatino 10 point Bold is
significantly higher than 9 point bold. However, 12 point Bold was not
significantly better than 10 or 9 point Bold.
Null Hypotheses 4 and 6 must be accepted. Number 5 is rejected.
Table 4 Sample A y Palatino Bold Type by Size
n=54
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 12 Bold
1.81
p=0.014
yes
Palatino 10 Bold
2.278
p=0.000
yes
Palatino 9 Bold
1.333
p=0.008
yes
Palatino 12 Bold
1.81
Table 5 Sample B y Palatino Bold Type by Size
n=53
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 12 Bold
2.226
p=0.430
no
Palatino 10 Bold
2.06
p=0.320
no
Palatino 9 Bold
2.264
p=0.830
no
Palatino 12 Bold
2.226
Table 6 Pooled Data y Palatino Bold Type by Size
n=107
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 12 Bold
2.019
p=0.300
no
Palatino 10 Bold
2.170
p=0.009
yes
Palatino 9 Bold
1.794
p=0.092
no
Palatino 12 Bold
2.019
Comparison of Palatino Italic Type by Size
Null Hypotheses (NH)
NH7: There is no significant difference between Palatino
Italic 12
point and Palatino Italic 10 point type viewed on
computer monitors.
NH8: There is no significant difference between Palatino
Italic 10
point and Palatino Italic 9 point type viewed on
computer monitors.
NH9: There is no significant difference between Palatino
Italic 12
point and Palatino Italic 9 point type viewed on
computer monitors.
Sample A and B show Palatino 12 point Italic significantly more
legible than Palatino 10 point Italic (Tables 7 and 8). Both samples show 9
point Italic to be significantly higher than 10 point Italic.
There is disagreement between samples on the comparison of 12 point
versus 9 point italic. Sample A shows 12 point to be significantly better while
Sample B shows 12 point to have a higher mean but not at a significant level.
Pooling the data shows 12 point Italic to be significantly better
than 9 point Italic (n=107; 12 point Italic mean = 1.800; 9 point Italic mean =
1.252; p=0.000).
Therefore, we must reject the null hypothoses in all three cases.
Table 7 Sample A Palatino Italic Type by Size
n=54
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 12 Italic
1.960
p=0.000
yes
Palatino 10 Italic
0.704
p=0.025
yes
Palatino 9 Italic
1.056
p=0.000
yes
Palatino 12 Italic
1.960
Table 8 Sample B Palatino Italic Type by Size
n=53
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 12 Italic
1.640
p=0.000
yes
Palatino 10 Italic
0.887
p=0.028
yes
Palatino 9 Italic
1.453
p=0.34
no
Palatino 12 Italic
1.640
Comparison of Palatino 12 Point Type by Face
Null Hypotheses (NH)
NH10: There is no significant difference between Palatino
12 point
and Palatino 12 point bold type viewed on computer
monitors.
NH11: There is no significant difference between Palatino
12 point
bold and Palatino 12 point Italic type viewed on
computer monitors.
NH12: There is no significant difference between Palatino
Italic
12 point and Palatino 12 point type viewed on
computer monitors.
Results are fairly consistent between Sample A and Sample B.
As Tables 9 and 10 indicate, there are significant legibility
differences in 12 point type based on differences in bold and italic face.
Sample A shows 12 point Normal to be significantly more legible than
the Bold face and the Italic face although there are no significant differences
between bold and italic.
Sample B also shows 12 point Normal to be significantly more legible
than Italic. Due to a higher mean for the Bold face, however, it is comparable
to the Normal face and significantly higher than Italic.
In this case, results will be pooled. A larger sample size will
help to generate a truer picture of the normal curve for each sample for t-test
comparisons.
Based on the pooled data (Table 11), Palatino 12 point is
significantly easier to read than 12 Bold or 12 Italic. Null Hypothoses 10 and
12 must be rejected and Null Hypothosis 11 accepted, that there is no
significant difference between Bold and Italic in 12 point Palatino.
Table 9 Sample A Palatino 12 point Type
n=54
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 12
2.352
p=0.005
yes
Palatino 12 Bold
1.810
p=0.46
no
Palatino 12 Italic
1.960
p=0.048
yes
Palatino 12
2.352
Table 10 Sample B Palatino 12 point Type
n=53
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 12
2.434
p=0.222
no
Palatino 12 Bold
2.226
p=0.003
yes
Palatino 12 Italic
1.640
p=0.000
yes
Palatino 12
2.434
Table 11 Pooled Data y Palatino 12 point Type
n=107
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 12
2.393
p=0.004
yes
Palatino 12 Bold
2.019
p=0.130
no
Palatino 12 Italic
1.800
p=0.000
yes
Palatino 12
2.393
Comparison of Palatino 10 Point Type by Face
Null Hypotheses (NH)
NH13: There is no significant difference between Palatino
10 point
and Palatino 10 point bold type viewed on computer
monitors.
NH14: There is no significant difference between Palatino
10 point
bold and Palatino 10 point Italic type viewed on
computer monitors
NH15: There is no significant difference between Palatino
Italic
10 point and Palatino 10 point type viewed on
computer monitors.
Results here are very consistent between Sample A and the
replication in Sample B. In both samples Palatino 10 bold had a higher mean
than 10 point normal. However, only Sample A showed a significant difference.
For both samples Palatino 10 point bold and 10 point normal were
significantly more legible than the italic face.
In this case, results will be pooled. A larger sample size will
help to generate a truer picture of the normal curve for t-test comparisons.
In each case the Null Hypothses must be rejected as there are
significant differences between groups. Palatino 10 Bold is more legible than
10 point Normal or Italic. Palatino 10 point Normal is more legible than 10
point Italic.
Table 12 Sample A y Palatino 10 point Type
n=54
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 10
1.310
p=0.000
yes
Palatino 10 Bold
2.278
p=0.000
yes
Palatino 10 Italic
0.704
p=0.001
yes
Palatino 10
1.310
Table 13 Sample B y Palatino 10 point Type
n=53
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 10
1.64
p=0.066
no
Palatino 10 Bold
2.06
p=0.000
yes
Palatino 10 Italic
0.887
p=0.001
yes
Palatino 10
1.640
Table 14 Pooled Data y Palatino 10 point Type
n=107
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 10
1.48
p=0.000
yes
Palatino 10 Bold
2.17
p=0.000
yes
Palatino 10 Italic
0.794
p=0.000
yes
Palatino 10
1.48
Comparison of Palatino 9 Point Type by Face
Null Hypotheses (NH)
NH16: There is no significant difference between Palatino 9
point
and Palatino 9 point bold type viewed on computer
monitors.
NH17: There is no significant difference between Palatino 9
point
bold and Palatino 9 point Italic type viewed on
computer monitors.
NH18: There is no significant difference between Palatino
Italic 9
point and Palatino 9 point type viewed on computer
monitors.
As Tables 15 and 16 indicate, there are inconsistent findings
between Sample A and Sample B. In Sample A there is no significant difference
between 9 point Normal and Bold. In Sample B there is a significant difference
with Bold being more legible. In Sample A there is no significant difference
between 9 point Bold and Italic. In Sample B, 9 point Bold scores significantly
higher than Italic.
For a comparison of Normal to Italic, the results are contradictory.
Sample A found Palatino Normal type to be significantly more readable than
Italic type in the 9 point sample. Sample B found 9 point Italic to have a
higher mean than Normal, but not at a significant level.
In this case, results will be pooled. A larger sample size will
help to generate a truer picture of the normal curve for each sample for t-test
comparisons.
The Null Hypothoses 16 and 17 must be rejected as Palatino 9 Bold
was more legible than 9 Normal and 9 Italic. We accept Null Hypothosis 18 that
there is no significant difference betwen 9 point Normal and Italic.
Table 15 Sample A y Palatino 9 point Type
n=54
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 9
1.480
p=0.43
no
Palatino 9 Bold
1.333
p=0.093
no
Palatino 9 Italic
1.056
p=0.028
yes
Palatino 9
1.480
Table 16 Sample B y Palatino 9 point Type
n=53
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 9
1.110
p=0.000
yes
Palatino 9 Bold
2.264
p=0.000
yes
Palatino 9 Italic
1.453
p=0.10
no
Palatino 9
1.110
Table 17 Sample B y Palatino 9 point Type
n=53
Type Style
Mean # words
T-test
p-value
Significant p=0.05
Palatino 9
1.300
p=0.001
yes
Palatino 9 Bold
1.794
p=0.000
yes
Palatino 9 Italic
1.252
p=0.74
no
Palatino 9
1.300
Rankings by Mean of Words Recognized
In Table 18 each type size and face is ranked by mean number of
words recognized from highest to lowest for each of the two samples.
TABLE 18 Rankings
Sample A
Sample B
Type Style
Mean # words
Mean # words
Palatino 12
2.350
Palatino 12
2.430
Palatino 10 Bold
2.280
Palatino 9 Bold
2.260
Palatino 12 Italic
1.960
Palatino 12 Bold
2.230
Palatino 12 Bold
1.820
Palatino 10 Bold
2.060
Palatino 9
1.480
Palatino 12 Italic
1.640
Palatino 9 Bold
1.330
Palatino 10
1.640
Palatino 10
1.320
Palatino 9 Italic
1.450
Palatino 9 Italic
1.060
Palatino 9
1.110
Palatino 10 Italic
0.700
Palatino 10 Italic
0.89
Note: For each sample (Sample A n=54; Sample B n=53) a difference
of approximately 0.38 to 0.40 generates a significant difference in the means
at the 0.05 level.
Since there were no significant differences for 8 of the 9 cells
between Sample A and Sample B pooling the data indicates an overall ranking for
the Palatino type faces and sizes. Caution should be used for the 9 point bold
figures as they did show significant differences between sample groups.
Table 19 Pooled Data for Palatino
n=107
Type Style
Mean # words
Palatino 12
2.393
Palatino 10 Bold
2.170
Palatino 12 Bold
2.025
Palatino 12 Italic
1.800
Palatino 9 Bold
1.795
Palatino 10
1.48
Palatino 9
1.295
Palatino 9 Italic
1.255
Palatino 10 Italic
0.795
Discussion
Note: For this section, Appendix One will be a useful reference as
it shows each type face reproduced at screen resolution and enlarged 200% to
show pixel detail.)
Palatino is a relatively light weight type face using thin, even
strokes for each letter. On the computer screen at 72 pixels per inch, this
translates to each stroke being one pixel wide. This remains consistent at 12,
10 and 9 point sizes. Therefore, the variability in size comes not in stroke
weight, but in the size of the letter forms through x-height and length of
ascenders and decenders.
For 12 point type (since the are 72 points per inch and 72 pixels
per inch a 12 point face will have 12 pixels height, 10 point 10 pixels, etc.)
the X-height is 6 pixels with ascenders 3 pixels and decenders 4 pixels. The
longer descenders give exceptional definition to the "g" and "j". In effect,
Palatino is "cheating" an extra pixel at 12 point and is actually 13 point type.
For 10 point, the X-height is 5, with ascenders and descenders each
having 3 pixels. Again, Palatino is "cheating" an extra pixel at 10 point and
is in effect 11 point type.
At 9 point with an X-height of only 5 pixels, letters hold their
form well and bowls do not fill in. Ascenders and descenders are 2 pixels each.
Two pixels is really the minimum for ascenders and decenders as dotted letters
need one pixel for the dot and one for the space. Losing an additional pixel
on the descenders would fill in the "g" below the line, however this could be
considered relatively minor since there are no other descenders with the round
descender form.
Palatino 12 has the highest mean number of words recognized and is
significantly better than 10 point and 9 point type. Palatino on screen has
very open bowls and holds the open spaces in letters such as "a" and "e" very
well even in smaller sizes. However, in 10 and 9 point sizes, the serifs begin
to merge between letters. This may in part explain the loss of legibility.
Palatino Bold gives mixed results. Palatino Bold appears to
significantly decrease legibility for 12 point although the two samples were
mixed in results. Bold may significantly increase legibility (samples differ)
for 10 point type and 9 point type. Since Palatino is a light face, using the
bold feature doubles the vertical stroke width to two pixels while leaving the
horizontal strokes at one pixel. The effect is to help smooth out thin or
jagged edges, especially on letters using slanting strokes such as "v" and "x"
and adds weight to rounded forms as well. One interesting note is that in 10
point Bold, some of the letters, notably "m" and "n" lose their serifs. They
re-appear on 9 point Bold.
Palatino Italic should be used with extreme caution. If used at
all, it should be used only at the 12 point size. In nearly all cases it was
significantly less legible than Normal and generally less legible than bold.
Using bold for emphasis would be much better as it scored significantly higher
in most samples. Another interesting observation is a comparison of 10 point
Italic with 9 point Italic when the type is enlarged. Ten point type is quite
slanted with a very jagged appearance as is typical for italic on the computer
screen. Nine point appears as if the designer drew a line horizontally across
the center of the letters and offset the letters one pixel so there is really
very little slant. This may account for the observation that 9 point Italic is
easier to read than 10 point.
As the final table (Table 19) indicates there are some clear breaks
in scores. Palatino 12 and Palatino 10 Bold are the most legible and have no
significant difference between them. There is a significant difference
between 12 and 12 Bold.
Using this as a break point, Palatino 12 Bold, 12 Italic and 9 Bold
are statistically not significantly different.
The next statistically similar grouping is Palatino 10 Normal,
Palatino 9 Normal and Paltino 9 Italic.
Palatino 10 Italic is in a grouping of its own with very low
legibility.
Findings from this study indicate:
1) There are significant differences for legibility of Palatino
12, 10 and 9 point type in the normal face at screen resolutions. Legibility
drops off significantly below 12 point.
2) Bold appears to give mixed results. Bold face significantly
decreases the legibility in 12 point and increases the legibility in 10 and 9
point. As a group however, it appears to strengthen the legibility. At all
sizes, it is a better choice for emphasizing material than italic.
3) Italic type should be avoided as it severly decreases
legibility, especially at small sizes.
Appendix One
Figure One y Palatino alphabets produced and copied at screen
resolution of 72 pixels per inch.
[--- Pict Graphic Goes Here ---]
Figure Two y Palatino alphabets produced and copied at screen
resolution of 72 pixels per inch and enlarged 200% to show pixel detail.
[--- Pict Graphic Goes Here ---]
[--- Pict Graphic Goes Here ---]
[--- Pict Graphic Goes Here ---]
SOURCES
Felici, J. (1987). Desktop Publishing Skills : A Primer for
Typesetting with Computers and Laser Printers. Reading MA : Addison-Wesley.
Felici, J. The Desktop Style Guide. Bantam, ITC.
Felici, J. (1996, March). A Bitmap a Day Keeps the Eyedoctor Away
[Internet Document]. Internet locator: http://www.will-haris.com/typofeli.htm.
McVey, G.F., & Weigeshaus, C.J. (1973) Studies in Legibility.
Rochester, NY: The Center for Visual Literacy, University of Rochester.
Petzold, C. (1992, November 24). Using text and fonts under the
OS/2 2.0 Presentation Manager. PC Magazine, Vol. 11, no.20. p. 453ff.
Tinke K. (1996, March/April). Taking It In. Adobe Magazine, pp.
40-45
Tinker, M.A. (1963). Legibility of Print. Ames IA: Iowa State
University Press
Will-Harris, D. (1996, March). The Best Faces For the Screen.
[Internet Document]. Internet locator: http://www.will-haris.com/typoscrn.htm.
Matrix News. (1995). Second TIC/MIDS Internet Demographic Survey,
[Internet Document]. Internet locator: http:www.mids.org.1995.