In summary, CVN subjects made errors on 2.2% of the colors on the RLLT (a total of 14 color misnamings, 30 missed colors, and 4 blanks named as a color by 46 subjects from a total of 48 stimuli). The colors that were not seen were mainly red, and this has been addressed in part 1 of this paper.1 The color misnamings were mainly naming red as yellow. This is a relatively common observation in practice11 although the reverse may occur with tinted lenses.12
On the RLLT, CVD subjects made errors on 37.2% of the colors (a total of 486 color misnamings, 156 missed colors, and 18 blanks called a color by 37 subjects from a total of 48 stimuli). The distribution of these errors, as a function of color named and color displayed, is also shown in Table 1. The errors of the CVD subjects are more varied compared with the CVN subjects.
The performance of the CVN subjects on the RLLT was used to establish a pass criterion based on the 95th percentile normal results. This was set as less than or equal to two misnamings, less than or equal to one green or yellow missed, and less than or equal to one blank named as a color. Any missed or misnamed reds constituted a failure given the greater safety implications of missing a red signal on the railways.
Fig. 1 also contains both CVN and CVD data for the CNLAN. Color vision–normal subjects made errors on 0.7% of the colors (10 misnamings and 2 missed colors by 46 subjects from a total of 39 stimuli). Color vision–deficient subjects made errors on 28.6% of the colors (407 misnamings and 5 missed colors by 37 subjects from a total of 39 stimuli). In accordance with Hovis and Oliphant,4 a single miscalling of yellow was not considered a fail, but more than one error is a fail. As a consequence, four CVN subjects failed the CNLAN (n = 1 miscalled green as yellow, n = 2 miscalled two yellows, and n = 1 miscalled three yellows).
It should be noted that a blank presentation is not a feature of the CNLAN; thus, the subjects knew that failure to see a light was an error. On the RLLT, there were blank presentations and, as a consequence, reporting a blank was preestablished as a legitimate response.
Fig. 2 shows the relationship between misnamings on the CNLAN and RLLT. The correlation for the CVD subjects and CVN subjects combined was 0.84 (n = 85, p < 0.01), and that for the CVD subjects alone was 0.57 (n = 37, p < 0.01). There was no significant correlation between the errors for the CVN subjects alone.
Given the pass/fail criteria for the two lanterns, the comparison of pass/fail performance was derived and is shown in Table 2. The data in Table 2 were analyzed using the Cohen kappa coefficient, κ,13 which is a measure of interrater agreement. The data, which include the six CVN subjects who failed the RLLT but who passed the CNLAN, give κ = 0.81. Values of κ that are greater than or equal to 0.81 are rated as “almost perfect agreement.”14 The predominant error was missing the two lowest luminance reds and the intensities of these lights were later modified, as described in part 1 of this study.1 If the values in Table 2 are amended, given the likely consequence of the modified intensities, and an assumption is made that, perhaps, only one CVN subject will fail the RLLT but pass the CNLAN, then κ = 0.95.
Subjects generally made more errors on the RLLT. The patterns of errors for the RLLT and CNLAN (Table 1) are rather different, mostly because the RLLT contains blank signals and reporting “no light” is an acceptable response. This was done partly because, in practice, there are not always two signals showing and partly to minimize the ability of protans to interpret red by its lowered luminous intensity. However, there was no evidence in the results that protans reported a blank more often than other subjects. If the data reporting blanks is removed, then there is a similarity in the misnaming of colors by CVD subjects in Table 1; that is, the most common error for both lanterns is the misnaming of yellow.
The reporting of blanks for red lights by CVN subjects was associated with the two lowest intensity red lights. Color vision–deficient subjects also frequently missed these reds but always made other errors, which, on their own, would have resulted in them failing the RLLT. A subsequent study was carried out on CVN subjects with the two lowest intensities doubled and 1 subject, out of 106, reported a blank for one red light and no subjects misnamed any colors.1 As a consequence, the production of RLLT includes the two modified intensities. Some or all of the six CVN subjects in this part of the study who failed the RLLT as a consequence of failing to see reds, but passed the CNLAN, might be expected to pass the modified RLLT (as manufactured). Missing a signal does occur for reasons other than color vision, such as attention and concentration. The subjects carried out a number of tests over an extended experiment, such that boredom and concentration lapses might well have intervened.13,14
In the validation study of the CNLAN,4 4% of CVN subjects failed and 3% of CVD subjects passed the CNLAN. In this study, the figures are 2% (1 in 46) and 5% (2 in 37), respectively. Given that the CVD population is unlikely to be exactly the same in both studies especially given that recruitment methods differ, these results are fully comparable. Given this “almost perfect agreement” of the CNLAN and RLLT shown by the Cohen κ, the two lanterns may be seen as being of essentially the same difficulty, despite their different design, signaling practice basis, and pass/fail criteria. Because they are both based on railway signaling practice, it seems reasonable to conclude that, again despite their differences, the difficulty in interpreting Canadian and New South Wales railway signals is also very similar.
If the results of this and the previous study4 are combined, the failure rate of CVD subjects is comparable with the reported fail rate of 98% of the Holmes-Wright Type B lantern.2 They are also more stringent than the Holmes-Wright Type A lantern (77% failure rate), Beyne Lantern (60%), OPTEC 900 (70%), and Farnsworth Lantern (66%), as summarized by the CIE.2 The Holmes-Wright Type A lantern has been identified as being insufficiently stringent for use on the railways.4
Despite differences in the signaling practice on which they are based, the RLLT and the CNLAN require very similar levels of color vision to pass. The Holmes-Wright Type B lantern has been identified as “a lantern test that presents a high level of difficulty” and is considered appropriate as a test used to assess applicants for CIE Color Vision Standard 1.2 The proportion of CVD subjects failing the Holmes-Wright Type B lantern (98%) is essentially the same as the RLLT and CNLAN; thus, it is also reasonable to consider the RLLT and CNLAN as appropriate for applicants for CIE Color Vision Standard 1. Unlike the Holmes-Wright Type B lantern, the RLLT and CNLAN are currently commercially available.
This study was supported, in part, by a grant from RailCorp NSW (now Sydney Trains). The assistance of Clair Taylor in data collection in experiment 1 is appreciated.
The CNLAN is available by contacting Dr. JK Hovis at the School of Optometry and Vision Science, University of Waterloo, Canada. The RLLT is available from ART Electronics ( www.rllt.com.au ). Sydney Trains own the rights to the RLLT.
Received January 31, 2014; accepted August 19, 2014.
1. Dain SJ, Casolin A, Long J. Color vision
and the railways: Part 1. The Railway LED Lantern Test. Optom Vis Sci 2015; 92: 138–46.
2. Commission Internationale de l’Éclairage. International Recommendations for Colour Vision Requirements for Transport
. Vienna, Austria: Commission Internationale de l’Éclairage; 2001.
3. Dain SJ, Casolin A, Long J. Color vision
and the railways: Part 3. Comparison of FaLant, OPTEC 900 and Railway LED lantern tests
. Optom Vis Sci 2015; 92: 152–6.
4. Hovis JK, Oliphant D. A lantern color vision
test for the rail industry. Am J Ind Med 2000; 38: 681–96.
5. Casolin A, Katalinic PL, Yuen GS, Dain SJ. The RailCorp Lantern test. Occup Med (Lond) 2011; 61: 171–7.
6. Williams CH. Standards of form and color-vision required in railway service. Trans Am Ophthalmol Soc 1897; 8: 227–41.
7. Williams CH. An improved lantern for testing color-perception. Trans Am Ophthalmol Soc 1900; 9: 192–6.
8. Edridge-Green FW. Two new tests for colour blindness. Brit Med J 1890; 1: 72.
9. Holmes JG, Wright WD. A new colour-perception lantern. Color Res Appl 1982; 7: 82–8.
10. Hovis JK, Oliphant D. Validity of the Holmes-Wright lantern as a color vision
test for the rail industry. Vision Res 1998; 38: 3487–91.
11. Wood JM, Atchison DA, Chaparro A. When red lights look yellow. Invest Ophthalmol Vis Sci 2005; 46: 4348–52.
12. Hovis JK. When yellow lights look red: tinted sunglasses on the railroads. Optom Vis Sci 2011; 88: 327–33.
13. Carletta J. Assessing agreement on classification tasks: the Kappa statistic. Comput Linguist 1996; 22: 249–54.
14. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977; 33: 159–74.
Keywords:© 2015 American Academy of Optometry
accident prevention; visual function; vision standard; medical standard; color vision; color-vision loss; color vision standards; railroad worker; transport; occupational risk; safety standard; lantern tests; practical test