Graefe’s Arch Clin Exp Ophthalmol (2007) 245:965–971

DOI 10.1007/s00417-006-0474-4

CLINICAL INVESTIGATION

The Freiburg Visual Acuity Test-Variability unchanged

by post-hoc re-analysis
Michael Bach

Received: 26 January 2006 / Revised: 28 August 2006 / Accepted: 9 October 2006 / Published online: 12 January 2007
# Springer-Verlag 2007

Abstract post-hoc analysis, for high reliability more trials should be

Background The Freiburg Visual Acuity and Contrast Test employed than the 18 trials per run used here.
(FrACT) has been further developed; it is now available for
Macintosh and Windows free of charge at http://www. Keywords Visual acuity . Automation . Computer
michaelbach.de/fract.html. The present study sought to
reduce the test-retest variability of visual acuity on short
runs (18 trials) by post-hoc re-analysis. Introduction
Methods The FrACT employs advanced computer graphics
to present Landolt Cs over the full range of visual acuity. It is surprising that the signal detection theory [9] and
The sequence of optotypes presented follows an adaptive computer assistance have not been widely exploited for the
staircase procedure, the Best-PEST algorithm. The Best- measurement of visual acuity (VA), although these methods
PEST threshold obtained after 18 trials was compared to the provide a considerable reduction of confounding influences.
result of a post-hoc re-analysis of the acquired data, where In contrast, clinical perimetry has extensively utilized these
both threshold and slope of the psychometric function were methods. In an attempt to fill this gap, the Freiburg Visual
estimated via a maximum-likelihood fit. Acuity and Contrast Test (FrACT) was developed [1]. The
Results Testing time was 1.7 min per run on average. Test- use of computer graphics and interactive computer assistance
retest reproducibility was ±2 lines (or ±0.2 logMAR) for a for the measurement of VA appeared quite straightforward.
95% confidence band (using 18 optotype presentations per However, unexpected hurdles appeared, including (1) suffi-
test run). Post-hoc psychometric fitting reproduced the cient resolution of display devices, (2) good automatic
Best-PEST result within 1%, although the individual slopes bracketing of the threshold, (3) problems of automatically
varied widely; test-retest reproducibility was not improved. dealing with the vagaries of patient responses, and (4)
Conclusions The FrACT offers advantages over traditional unrealistic expectations of the stability of acuity measures,
chart testing with respect to objectivity and reliability. The even if obtained quasi-objectively. Since then, FrACT has
similarity between the results of the Best-PEST vs. post-hoc been used in a number of studies, while being continuously
analysis, fitting both slope and threshold, suggest that there further developed based on the feedback of numerous vision
is no disadvantage to the constant slope assumed by Best researchers, optometrists and ophthalmologists. Other
PEST. Furthermore, since variability was not reduced by approaches emulated the ETDRS charts [18] or were variants
of the FrACT approach [16]. ETDRS and FrACT results
were recently compared in this laboratory and found to agree
closely [19]. The present article will describe the current
version of the FrACT and a test of the viability of the
M. Bach (*) constant slope assumption of the Best-PEST threshold
University Augenklinik Freiburg,
estimation paradigm. It was hoped that post-hoc processing
Killianstr. 6,
79106 Freiburg, Germany could reduce the test-retest variability, so that briefer test
e-mail: [email protected] runs could be employed.
966 Graefe’s Arch Clin Exp Ophthalmol (2007) 245:965–971

Fig. 1 Landolt C. The unit u, are very low relative to current computer capabilities. The
measured in minutes of arc, graphics should be able to resolve at least 256 gray levels,
defines the decimal visual acuity
(VA): For a VA of 1.0 (=20/20), or millions of colors (3×8-bit color depth). The resolution
u would span 1 min of visual of the visual display unit (VDU) is the most likely
angle 1u bottleneck (see “Limitations” in Discussion). Both CRT-
or LCD-type VDUs are possible. EN ISO 8596 [6] details a
luminance of the Landolt C between 80 and 320 cd/m2 at a
contrast of 85%. This is easily reached with consumer-
5u grade equipment. The present results were obtained with a
17” CRT monitor at a distance of 4 m, a luminance of
Methods 105 cd/m2 at a contrast of 95% and a background
illuminance of 60 lux, measured at the subject’s eye.
Versions

FrACT is currently available in two different versions: (1)

the old version, which runs only on Macintosh computers, Nomenclature and optotypes
but also offers contrast testing, and (2) the universal
version, the latter being programmed such that executables Visual acuity will be abbreviated with VA and defined
for two major computer platforms are produced (Macintosh operationally; the decimal notation will be used throughout
and Windows) and having a web-browser plug-in that also [FrACT results are also optionally displayed as Snellen
runs on Linux. The two versions are identical with respect ratio or logMAR = −log(VA)]. The operational definition of
to the optotype geometry, calibration and threshold estima- VA is as follows: VA=1/dt where dt= the threshold gap size
tion algorithm; they differ in the programming environment (minutes of arc). The definition of threshold will be given
and optional capabilities. The present report concentrates in the next section. The term gap refers to the gap of the
on acuity assessment with the second, more recent, Landolt C (Fig. 1). Landolt C optotypes are presented on
platform-agnostic version. the computer screen. The relation of the outer and inner
diameter to the gap size is shown in Fig. 1. The
Equipment independent variable is the gap size. Resolution is im-
proved by using anti-aliasing [2, 8], allowing sub-pixel
Nearly any kind of computer manufactured in the last rendering of the Landolt C, and for non-integer gap sizes
5 years can be employed. The computational requirements (e.g., 1.25 pixels). The gap can be presented in one of eight

Fig. 2 The psychometric func- 100%

tion governing acuity measure-
ments, here shown for a subject
with decimal VA=0.61. When
the optotypes are large (left),
Probability of correct report

they will be recognized correct-

ly. With decreasing optotype
size, the probability to report the
gap direction correctly will de-
crease, eventually down to the Inflection point
guessing rate, which equals
= 56.25%
12.5% for eight possible gap 50%
directions. In between, the de-
tection rate is well described by
a logistic function. The acuity is
defined at the inflection point,
the middle between 100% cor-
rect and the guessing rate. For
subjects with differing acuity, Guessing rate = 12.5%
this psychometric function shifts
horizontally; the dashed curve
example corresponds to
VA=0.3 0%
0.1 0.3 1.0 3.0
VA = 0.61
Optotype size in acuity units
Graefe’s Arch Clin Exp Ophthalmol (2007) 245:965–971 967

or in one of four positions; in the present study, eight value of the gap size ≥1 pixel is possible though, with the
different orientations were used. help of anti-aliasing. The Landolt C orientation is calculat-
ed randomly for each trial.
Threshold definition An important parameter is the number of trials [13]. In
the interest of rapid measurement and avoidance of subject
Like all sensory thresholds, the detection rate versus fatigue, it should be as low as possible; for best precision,
optotype size is described by a psychometric function however, it should be as high as possible. For clinical
(Fig. 2) [11]. Given this continuous relation, threshold studies, 30 trials yield a test-retest comparable to the
definition is not obvious. The optimal choice from a signal- ETDRS procedure [19]. In the present study, only 18 trials
detection point of view is the point of steepest slope, which were presented. This increases variability, making room for
is also the point of inflection. With the use of eight improvement. Post-hoc re-analysis (see below) sought to
alternatives, this point lies in the middle between the reduce the variability, though without success.
guessing rate of 12.5% and 100%, i.e., at 56.25%. At this
point, any deviation on the detection scale (ordinate) Procedure
transforms into the smallest possible deviation on the
acuity scale (abscissa). This definition is widely used and The procedure with FrACT is subject to the same boundary
also underlies the EN ISO 8596 standard [6]. One could conditions as any acuity test: well-defined surround lumi-
also define this region as the most uncomfortable one for nance, no screen illumination that would reduce contrast, a
the patient: here, they are most uncertain whether or not
they can recognize the target.
3.0
Threshold estimation ø
Gap size [acuity units]

ø ø xø xøx xø
It is suggested by signal detection theory [9], and has been 1.0 x x x
shown experimentally (e.g., [17]), that when comparing the
psychometric acuity function in subjects with a wide range x x x
0.3
of acuity, the position of the inflection point shifts, but x
slope stays rather constant, when plotted on a log(VA) 0.1 x
scale. Neglecting lapses, only one parameter needs to be
estimated, namely the threshold. A number of algorithms #61, VA=1.28
have been developed for such a situation (review: [20] or 0.03
the special edition of Perception [15]). For FrACT, the 0 5 10 15 Trial #
Best-PEST algorithm was selected [12], which needs no
prior information and assumes a constant fixed slope of the
3.0
Gap size [acuity units]

psychometric function. In its purest form, this algorithm

always presents optotype sizes at the currently most likely 1.0
threshold, because that maximizes information gain. This ø
x øø
most likely position is calculated by a maximum likelihood ø x øø
procedure that looks for the position of the psychometric
0.3 x x xøxø
x
function that would reproduce all previous trial results with
the highest likelihood. 0.1 x x x
Practically, a number of modifications are useful. The #59, VA=0.33
initial optotype size depends on the highest and lowest 0.03
possible acuity (the full range); thus, it would be strongly 0 5 10 15 Trial #
influenced by screen size, pixel size and observation Fig. 3 Two representative runs of FrACT in two subjects. The
distance. To avoid such influences, the first four trials ordinate represents the Landolt-C gap size; the abscissa covers the
sequence of 18 trials. Trials with correct responses are indicated by
,
present optotypes corresponding to an acuity of 0.1, 0.2, 0.4
incorrect ones by ∅. The low-acuity outliers at trial 12 and 18
and 0.8 (following EN ISO 8596 [6]) as long as they are all represent the bonus trials. In the top subject, the initial 0.1–0.8 acuity
correctly discriminated. From then on, the Best PEST takes sequence was recognized correctly, but the presentation at nearly
over. As a boundary condition, the optotype never becomes 3.0 VA was not correctly identified; the algorithm subsequently homed
in on VA=1.28. In the bottom subject, the Landolt C corresponding to
smaller than 5 pixels in diameter (corresponding to a gap 0.4 acuity was not correctly identified with a subsequent step-down in
size of 1 pixel), as—even with anti-aliasing—the effective acuity, with: ν, acuity value of the optotype; pchance, guessing
contrast would vanish below this limit. Any non-integer probability =0.125
968 Graefe’s Arch Clin Exp Ophthalmol (2007) 245:965–971

Acuity Post-hoc re-analysis 3.0

n = 148 (these results were discarded), then four runs in the
sequence left eye/right/right/left were recorded; for every
alternate subject, the sequence of eyes was inverted. The
data of every run were saved using the “export to
clipboard” option (see manual [3]). To extend the range of
1.0 acuities to lower values, some subjects were assessed in a
second session, about 2 weeks later, with glasses degrading
optical quality. The optical quality was not reduced with
plus lenses because it is likely that accommodation will be
unstable under such a condition. The commercially avail-
0.3 able Bangerter transparencies proved too unequal across
their surface. Prompted by a personal communication by
Bernhard Rassow, I experimented with plastic transparen-
cies manufactured as office paper binders until one was
found with an even distribution of its ‘milky’ quality and
0.1 the right amount of degradation, and mounted cutouts in a
0.1 0.3 1.0 3.0
cardboard trial frame. Such a scatter transparency, due to its
high amount of wide-angle scatter, models some media
Acuity bestPEST opacities [10].
Fig. 4 Comparison of acuity results obtained by Best-PEST and post-
hoc fitting of a psychometric function with slope and threshold as free
Most subjects can perform the test in a self-paced mode,
parameters especially on repeat visits. It is useful to help with
occasional supportive statements indicating that “errors”
calibrated setup, defined observer distance, appropriate are entirely in order, and not to ponder the response too
correction and adequate training of examiner and subject. long. This self-paced mode can allow the examiner to
To obtain data for the present study, FrACT was run in pursue other tasks in the mean time.
74 eyes of 37 subjects (age range, 19–71 years, mean
38 years) without known eye disease, using eight different Post-hoc processing
optotype orientations. Most performed the test for the first
time. Their correction was not necessarily the best, which is Since inception of this program, computer speed has risen
not a problem here, since test-retest variability was the by several orders of magnitude. Thus, the psychometric
target variable. The subjects operated a remote response paradigm needs no longer be chosen for speed, which had
box, with keys labeled by Landolt Cs with appropriate been a major design goal for the Best-PEST algorithm [12],
orientations. The test was explained, one binocular run and can be re-evaluated. Best PEST assumes a constant
performed to familiarize the subject with the procedure slope and zero lapse rate.

50 50 50

#144 #92 #0
Slope

Slope

Slope

40 40 40

30 30 30

20 20 20

10 10 10

0 0 0
0.3 1.0 3.0 0.3 1.0 3.0 0.3 1.0 3.0
Decimal acuity Decimal acuity Decimal acuity

Fig. 5 Typical characteristics of the psychometrical function (cf. acuity cases; the right one depicts a run where the subject gave an
Fig. 2) for three different subjects. The z-axis (represented via contour incorrect response to a (very easy) bonus trial, resulting in a low slope.
lines) depicts the likelihood of the fit producing the specific test run In none of the 148 cases was there a marked obliqueness of the
result; the z-ranges covered are: 0–5·10−3, 0–3·10−3, 0–6·10−5, likelihood ‘hill’ (the center graph represents an extreme case),
respectively. The abscissa represents the acuity threshold, and slope indicating that the threshold estimate does not depend strongly on
is on the ordinate.The two left graphs are typical for low and high the slope
Graefe’s Arch Clin Exp Ophthalmol (2007) 245:965–971 969

The run data thus obtained (similar to those depicted in has already occurred will yield a specific outcome. The
Fig. 3) were fitted with a psychometric function P using concept differs from that of a probability in that a
maximum-likelihood [21, 24] where both the threshold v0 probability refers to the occurrence of future events, while
and slope s were free parameters according to the following a likelihood refers to past events with known outcomes
Eq. (1):
Pðν Þ¼pchance þð1  pchance Þ=ð1þðν 0 =ν Þs ÞÞ ð1Þ a
3.0

Decimal acuity, retest

n = 74
To asses the quality of these post-hoc acuity values, their
test-retest variability was calculated. To quantify test-retest
variability, often the correlation coefficient is calculated.
This, however, depends strongly on the total range, and one 1.0
can achieve high values just by including endpoint values.
The mean coefficient of variation (CV) across every test-
retest pair does not suffer from this limitation and was
chosen instead. There is one problem using the CV though:
VA is not normally distributed, but logVA or logMAR are.
0.3
But the CV calculation cannot be applied to data crossing
zero, which is the case with both logVA and logMAR.
Since the test-retest values are rather close (around 10%), best PEST
and, as mathematicians say, everything is linear to the first mean CV = 12.9±9.7%

order, the CV was calculated on the VA scale. post-hoc

mean CV = 13.4%±9.6%
0.1
0.1 0.3 1.0 3.0
st
Decimal acuity, 1 test
Results b
Difference of test-retest [log(VA) = – logMAR]

0.4
In Fig. 3, two representative runs of FrACT in two subjects
are depicted. It can be seen that a run starts with “easy”
optotypes (low acuity) that become smaller until an
incorrect response is encountered (the fifth trial for the 0.2
upper example). Consequently, the Best-PEST algorithm Mean+2·SD
next selects a somewhat larger optotype.
The average time per 18-trial run, including data transfer
to a spreadsheet, was 103±40 s (range, 53–210 s). It took
0.0
about 6 min for the acuity of both eyes including a Mean
binocular training run.
Figure 4 demonstrates that the threshold obtained by
fitting the psychometric function with slope as another free
parameter never differed by more than one line (1 dB, or a -0.2
Mean–2·SD
factor of 1.26) from the one obtained by Best PEST. On
average, the results differed by 1.1%. The slope ranged
widely from 2.0 to over 100, averaging at 17.0; this is
markedly higher than the slope value of 1.7 currently used. -0.4
Figure 5 depicts the maximal likelihood function depending -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2
on slope and threshold (decimal acuity) for three represen- Mean of test-retest [log(VA) = – logMAR]

tative cases. On the left, the slope of the psychometric Fig. 6 Test-retest variability of FrACT using 18 trials: 74 eyes of 37
naive, visually normal subjects, not necessarily wearing best correc-
function is very steep, while on the right it is very shallow tion, were analyzed. (a) Scatter plot. Along the continuous 45° line,
(in that run, an incorrect response to a large optotype, a perfect reproducibility would be obtained. The parallel dashed lines
bonus trial, was entered). The threshold depends very little indicate a deviation of ±3 lines on retest from the initial test. There
on slope as can be seen from the missing obliqueness of the was no improvement of repeatability by post-hoc analysis. (b) The
best-PEST data, depicted as difference test-retest vs. average of test-
likelihood vs. slope and threshold function. A brief retest (Bland-Altman plot [4, 5]); the abscissa covers the same range
explanation of the likelihood function may be in order: as (a). The mean test-retest difference is close to zero, and variability
Likelihood is the hypothetical probability that an event that does not change markedly across the acuity range covered
970 Graefe’s Arch Clin Exp Ophthalmol (2007) 245:965–971

[22]. The event here is the occurrence of the entire reflects inherent fluctuation of the threshold itself, or the
sequence of correct-incorrect responses, given the specific loss of degrees of freedom to estimate the additional
values of acuity threshold and slope value (cf. Fig. 1). parameter slope offsets a possible closer approximation of
Figure 6 illustrates the test-retest reproducibility. Points the psychometric function. Still, post-hoc processing has
on the continous 45° line would be perfectly reproduced; been integrated into FrACT as an option.
the dashed lines correspond to deviations by a factor of two A number of additional modifications of the post-hoc
(corresponding to three lines on an acuity chart or 3 dB). analysis were tried out: removal of the bonus trial results,
For the Best-PEST method, 72 of 74 (97.3%) run pairs restricting analysis to the final part, iteratively removing
differed by 2 dB or less; the mean CV was 12.9±9.7%. outliers, and removing erroneous bonus trials. None of
After post-hoc processing, 73 of 74 (98.6%) run pairs these modifications resulted in a lower test-retest variability.
differed by 2 dB or less; the mean CV was 13.5±9.7%. One problem of FrACT occurs when a subject mistypes
Thus, post-processing does not yield significantly different their first response; that is when a very large optotype is
thresholds (P=0.79, Wilcoxon test), it does not reduce test- seemingly not recognized correctly. The Best-PEST algo-
retest variability, and it removes about as many outliers as it rithm then searches too long for the threshold in the low
adds. In Fig. 6b, the Best-Pest results are depicted as a acuity region and may not converge to full acuity. In such a
Bland-Altman [4, 5] plot (after taking the logarithm of all case, it is best to abort the run and restart. This was not
acuities, the difference test-retest vs. average of test-retest). necessary in the present study.
The mean difference (dotted line) was 0.025 logMAR, and The main application fields for FrACT are thus clinical
the dashed lines indicate ±2·SD around the mean (2·SD= studies where acuity is an outcome variable. FrACT can be
0.196 logMAR). This plot shows that: (1) The negative seen as an automated alternative to ETDRS, extending its
mean difference hints at a small, though insignificant (P= range both at the upper and lower end and being safe from
0.6) learning effect, and (2) there is no marked skewness for being learned by heart on repeated testing. In laboratory
low acuities, but a tendency towards higher variability. The environments, FrACT has proven useful for subject
post-hoc data have been left out to avoid clutter; the mean screening and for quantifying acuity after optical or
difference was 0.014 logMAR and the 95% confidence physiological manipulations. Since the present study was
band is spanned by 0.204 logMAR. not successful in reducing the variability of the rather short
18-trial runs, for highly reliable results, the test should
either be repeated and the results averaged, or the number
Discussion of trials should be increased to 30 [1, 19].

The old version of FrACT has been successfully validated Acknowledgement Thanks to many users for their inspiring support,
in independent laboratories [7, 14, 23]. The new version is providing feedback that helped to root out bugs and suggesting useful
expansions. Special thanks to Lew Harvey, Hans Strasburger and Thomas
geometrically identical and showed an agreement between
Meigen for tutoring in signal detection theory, psychometric threshold
the (new) FrACT and ETDRS charts within 9% down to assessment and probability statistics and to Margret Schumacher for
very low acuities in the author’s laboratory [19]. This assiduous testing. Finally, thanks to two very persistent reviewers who
suggests that the FrACT results are bias-free estimators of considerably helped to clarify my thoughts.
visual acuity over the full range from ≈0.01 to ≈3.0. The
present study assessed the test-retest variability of FrACT
with only 18 trials. The test-retest variability as quantified References
by the coefficient of variation (CV) of VA was around 13%,
corresponding to about half a line (1 line = a factor of 1.26). 1. Bach M (1996) The Freiburg Visual Acuity Test-automatic
The corresponding 95% confidence interval spans ±0.196 measurement of visual acuity. Optom Vis Sci 73:49–53
2. Bach M (1997) Anti-aliasing and dithering in the Freiburg Visual
logMAR. This leaves room for improvement. Acuity Test. Spat Vis 11:85–89
The post-hoc analysis, namely fitting slope as another 3. Bach M (2006) Homepage of the Freiburg Visual Acuity and
free parameter in addition to the threshold, resulted in Contrast Test (‘FrACT’). Retrieved 2006-07-04, from http://www.
nearly identical acuity estimates and nearly identical michaelbach.de/fract.html
4. Bland JM, Altman DG (1986) Statistical methods for assessing
average test-retest values. The maximum likelihood fitting agreement between two methods of clinical measurement. Lancet
surface showed that slope and threshold are highly 1:307–310
decoupled; in other words, whichever value of slope is 5. Bland JM, Altman DG (1995) Comparing methods of measure-
chosen has very little influence on the acuity outcome. This ment: why plotting difference against standard method is
misleading. Lancet 346:1085–1087
suggests that the fixed slope as used in Best PEST is no 6. CEN (Comité Européen de Normalisation) (1996) Ophthalmic
disadvantage. Somewhat disappointingly, post-hoc process- optics-visual acuity testing-the standard optotype and its presen-
ing did not improve test-retest variability. Either this tation. Beuth-Verlag, Berlin
Graefe’s Arch Clin Exp Ophthalmol (2007) 245:965–971 971

7. Dennis RJ, Beer JM, Baldwin JB, Ivan DJ, Lorusso FJ, Thompson 16. Peters BT, Bloomberg JJ (2005) Dynamic visual acuity using
WT (2004) Using the Freiburg Acuity and Contrast Test to “far” and “near” targets. Acta Otolaryngol 125:353–357
measure visual performance in USAF personnel after PRK. 17. Petersen J (1990) Zur Fehlerbreite der subjektiven Visusmessung.
Optom Vis Sci 81:516–524 Fortschr Ophthalmol 87:604–608
8. Foley JD, Van Dam A, Feiner SK, Hughes JF (1990) Computer 18. Ruamviboonsuk P, Tiensuwan M, Kunawut C, Masayaanon P
Graphics, Principles and Practice. Addison-Wesley, Reading (2003) Repeatability of an automated Landolt C test, compared
9. Green DM, Swets JA (1966) Signal detection theory and with the early treatment of diabetic retinopathy study (ETDRS)
psychophysics. Wiley, New York chart testing. Am J Ophthalmol 136:662–669
10. Hess R, Woo G (1978) Vision through cataracts. Invest Ophthalmol 19. Schulze-Bonsel K, Feltgen N, Burau H, Hansen LL, Bach M
Vis Sci 17:428–435 (2006) Visual acuities “Hand Motion” and “Counting Fingers”
11. Klein SA (2001) Measuring, estimating, and understanding the can be quantified using the Freiburg Visual Acuity Test. Invest
psychometric function: a commentary. Percept Psychophys Ophthalmol Vis Sci [in print]
63:1421–1455 20. Treutwein B (1995) Adaptive psychophysical procedures. Vision
12. Lieberman HR, Pentland AP (1982) Microcomputer-based esti- Res 35:2503–2522
mation of psychophysical thresholds: The best PEST. Behav Res 21. Treutwein B, Strasburger H (1999) Fitting the psychometric
Methods Instrument 14:21–25 function. Percept Psychophys 61:87–106
13. Linschoten MR, Harvey LO, Jr., Eller PM, Jafek BW (2001) 22. Weisstein EW. (2006) “Likelihood.” From MathWorld - A
Fast and accurate measurement of taste and smell thresholds Wolfram Web Resource. Retrieved 2006-06-27, from < http://
using a maximum-likelihood adaptive staircase procedure. mathworld.wolfram.com/Likelihood.html >
Percept Psychophys 63:1330–1347 23. Wesemann W (2002) [Visual acuity measured via the Freiburg
14. Loumann Knudsen L (2003) Visual acuity testing in diabetic visual acuity test (FVT), Bailey Lovie chart and Landolt Ring
subjects: the decimal progression chart versus the Freiburg visual chart]. Klin Monatsbl Augenheilkd 219:660–667
acuity test. Graefes Arch Clin Exp Ophthalmol 241:615–618 24. Wichmann FA, Hill NJ (2001) The psychometric function: I.
15. Macmillan NA (2001) Threshold estimation: the state of the art. Fitting, sampling, and goodness of fit. Percept Psychophys
Percept Psychophys 63:1277–1278 63:1293–1313