Acta Pdiatr 93: 830835.

2004

A comparison of Likert scale and visual analogue scales as response options in childrens questionnaires
H van Laerhoven, HJ van der Zaag-Loonen and BHF Derkx
Emma Childrens Hospital, Academic Medical Centre (AMC), University of Amsterdam, Amsterdam, The Netherlands

Laerhoven H, van der Zaag-Loonen HJ, Derkx BHF. A comparison of Likert scale and visual analogue scales as response options in childrens questionnaires. Acta Pdiatr 2004; 93: 830835. Stockholm. ISSN 0803-5253 Aim: To examine which response options children prefer and which they nd easiest to use, and to study the relative reliability of the different response options. Methods: A consecutive group of unselected children (n = 120) lled out three questionnaires in a paediatric outpatient clinic. Each questionnaire included seven similar questions, but had different response options: the Likert scale, the Visual Analogue Scale (VAS) and the numeric VAS. In general, the questions were not related to the childrens particular diseases, but dealt with the frequency of simple activities, their feelings and opinions. The pages with the three different response options were offered in random order. Afterwards, the children rated their preference and ease of use of the different response options on a scale from one to 10. Results: Children preferred the Likert scale (median mark 9.0) over the numeric VAS (median mark 8.0) and the simple VAS (median 6.0). They considered the Likert scale easiest to ll out (median mark 10 vs 9 and 7.5 for the numeric and simple VAS, respectively). Results of the different response options correlated strongly with each other (rho = 0.67 0.90, p < 0.05). Conclusion: Children prefer the Likert scale over the numeric and simple VAS and nd it easiest to complete. The Likert scale, the simple VAS and the numeric VAS are of comparable reliability. The Likert scale is recommended for use in questionnaires for children, although research into larger and more diverse samples is needed. Key words: Children, methodology, questionnaires, rating scales, response options HHF Derkx, PO Box 22660, 1100 DD Amsterdam, The Netherlands (Tel. 31 20 5669111, fax. 31 20 6917735, e-mail. h.h.derkx@amc.uva.nl)

There is an increasing interest in measuring emotional states and quality of life of children, both in clinical practice as well as in empirical research (1). For these measurements, questionnaires including several types of response options can be used. Of these response options, the Likert scale (verbal categorical response options), the simple Visual Analogue Scale (VAS) (line response option) and the numeric VAS (numeric response option) are used most frequently. Little is known about childrens preference for response options; most research focuses on questionnaires designed for adults. The aim of this study is to examine which of the three response options (Likert scale, simple VAS and numeric VAS) is preferred and which is easiest to use by children in different age groups. In questionnaires designed for adults, both the Likert scale and the VAS are applied, most often to quantify pain severity. They have proven to be convenient and valid quantication instruments (26). For the assessment of other outcomes, such as quality of life, the Likert response option is most often applied because of the few cognitive demands it places on respondents (easy to complete) and because it makes easy score
2004 Taylor & Francis. ISSN 0803-5253

computation possible for the investigator (610). Most authors agree that there are few differences in reliability and responsiveness between the response options, and they prefer the Likert scale because it is easy to interpret (2, 46, 8, 9, 11). However, research into adults cannot simply be transposed to the paediatric population, since children have different cognitive capacities. Generally, children from the age of 8 y onwards are believed to be able to provide reliable reports on their well being (12, 13). Some authors favour a standardized reading test instead of an age limit to establish the minimum skills required to complete questionnaires (13, 14). Although there is little empirical evidence to show the adeptness of children of different ages to choose between the different response options, the Likert response option is most often used in paediatric questionnaires (1). March et al. (15) found that younger children and children with poorer reading skills were less able to respond to negative items on questionnaires, the effect of which biased the interpretation of the childrens responses. As for the research on severity of pain in children, the VAS is considered to be the best response
DOI 10.1080/08035250410026572

ACTA PDIATR 93 (2004)

Response options in childrens questionnaires

831

option. It does not force the respondent into xed categories as does the Likert scale (16). Abu-Saad et al. (17) described in their research on pain experience in 355 children that the VAS has proven to be easy, effective and sensitive. Rebok et al. (13) also found that children, regardless of age, preferred circled response options to the VAS line in pain research, but because of the small study sample this conclusion is tentative (13). Theoretically, younger children who have less reading potential may be expected to prefer the VAS response option, because it requires less reading time and skills than the Likert type of response options. We studied the preference and ease of completing response options (Likert scale, simple VAS and numeric VAS), and used the correlation as a measure of relative reliability. We hypothesized, based on the literature and theory available, that younger children would prefer the VAS, whereas older children would prefer the Likerttype response option. In accordance with the studies into adult response options, we expected that the correlation between the different types of response options would be good.

mood) and opinions (about school, sports and height) (Fig. 1a, b and c). Afterwards, children were asked to rate how much they liked each answering option and how difcult they considered it to be answered, by giving them a mark from one (less preferred or most difcult) to 10 (most preferred or easiest).

Statistical analysis To assess the inuence of age on preference for response options, data were analysed in two subgroups: children 6 to 12 y old and children 13 to 18 y old. Differences between the two age groups relating to preference and difculty of the response options were tested with non-parametric procedures (Mann-Whitney U-test). Differences in time needed for completion were also tested with the Mann-Whitney U-test. Marks given for preference and difculty of each response option were treated as ordinal data. Therefore, median scores were calculated and non-parametric procedures were used. Differences in median preference and difculty scores within patients for the three response options were compared using the Wilcoxon statistic for paired ordinal variables. Differences in preference and difculty between genders and between immigrants and native children were analysed using the Mann-Whitney U-test. The reliability of each response option could not be established in this study as it is of cross-sectional nature. Instead, we chose to analyse the relative reliability of the three response options by computing Spearman rank order correlation coefcients between the various response options. The assumption was that at least one of the response options would best represent the true answer of each child. Therefore, a high correlation coefcient would represent high reliability of all three response options; likewise a low correlation coefcient would indicate that one (or two) response options were less reliable. To discover the potential systematic biases in the location of the answers (e.g. the outer answer options on the VAS line), we compared the percentage of children scoring in each category of the ve-point Likert scale with the percentage of children scoring in ve categories of the VAS. In case of absence of systematic bias, these percentages would have to be comparable. We therefore categorized the numeric VAS score range of 110 into ve even categories (scores 1 and 2 into category 1, 3 and 4 into category 2, etc.). Similarly, the line of the simple VAS was divided into ve equal parts. Wilcoxons rank statistic for paired ordinal data was used to test whether the number of children scoring in each category was consistent across response options. A conventional signicance level of p < 0.05 was chosen.

Methods
A consecutive group of 120 unselected children between 6 and 18 y old completed a survey containing three similar questionnaires with different response options. The survey was carried out during the childrens visit to a paediatric outpatient clinic and prior to consultation. Data registered concerned age, sex, type of school the children attended and ethnic background, as well as the time needed to complete the questionnaires. Ethnic background was dened as immigrants (rst, second or third degree non-native children) and native Dutch children. Each questionnaire contained a different response scale: a ve-point Likert scale (Fig. 1a), a simple VAS (Fig. 1b) and a numeric VAS (Fig. 1c). The simple VAS was a 10-cm line with the extreme answers of the Likert scale marking the ends. The children were asked to mark the line with a cross somewhere between both extremes that best reected their answer. The numeric VAS consisted of a series of numbers from one to 10 reecting the answering options, with again the extreme answers of the Likert scale at the beginning and the end of the series. The pages were presented to the children in random order. The children, who completed the questionnaires by themselves, were told that the same seven questions would be asked on three different pages. The investigator prevented the children from peeking at their previous answers and recorded time of completion. The seven-item questionnaire covered three domains: frequency of simple activities (watching television, using the bus), feelings (about dreams and current

832

H van Laerhoven et al.

ACTA PDIATR 93 (2004)

Fig. 1.

ACTA PDIATR 93 (2004)

Response options in childrens questionnaires

833

Table 1. Median marks for different response options.

Median mark Ease Preference Likert scale simple VAS numeric VAS Likert scale simple VAS numeric VAS 10.0a 7.5b 9.0 9.0c 6.0d 8.0 Age group 612 y 9.1 7.3 8.1 8.5 6.2 7.6 Age group 1318 y 9.2 7.2 8.0 8.3 6.2 7.6 Mann-Whitney U-teste p-value 0.10 0.89 0.68 0.21 0.81 0.61

Higher than the mark for the simple VAS (p < 0.001) and numeric VAS (p = 0.005). Lower than the mark for the numeric VAS (p < 0.001). c Higher than the mark for the simple VAS (p < 0.001) and numeric VAS (p < 0.001). d Lower than the mark for the numeric VAS (p < 0.001). e Differences in median marks for the three response options between the two age groups (children 612 and 1318 y old).
a b

Results
One hundred and twenty-two consecutive children were asked to participate. Two refused. This resulted in a response rate of 99%. Sixty-nine children (58%) were aged between 6 and 12 y, and 51 children (42%) were aged between 13 and 18 y; 57 children (48%) were boys and 53 children (44%) were immigrants. There were more instances of missed questions on the simple and the numeric VAS than on the Likert scale (3%, 2% and 0.5%, respectively). The questionnaires were completed in 2 to 22 min, where 75% of children needed between 5 and 10 min (mean 7.8 min). Younger children needed more time to complete the questionnaire: 8.5 min vs 7.0 m, respectively (p = 0.02). Immigrant children needed more time to complete the questionnaires: 8.8 min vs 7.2 min for native children (p = 0.05). Table 1 shows the median marks for preference and difculty of the different response options. The Likert scale scored signicantly higher on both preference and difculty, followed by the numeric VAS and the VAS (all p-values <0.005; Table 1). There was no difference in preference or difculty scores between the younger and older children (Table 1) or between boys and girls.

Immigrant children liked the simple VAS better than did native children (median score 7.0 vs 6.0, p = 0.003), and they found the simple VAS easier to ll out (9.0 vs 7.0, p = 0.001). Only 10 children attended a special school, and this group was considered too small to take into consideration. Table 2 shows the comparison of the percentage of children scoring in each category of the Likert scale and in each of the categories of the simple VAS. There appeared to be a systematic bias in the answers of the children, with a very high percentage of children consistently scoring in the rst (left end) category of the VAS. We therefore rearranged the VAS into ve new categories: the two most extreme categories were comprised (incorporating only 1 cm instead of 2), and the remaining 8 cm were divided into the remaining three categories, resulting in an uneven categorization of the VAS. With this new categorization, children still tended to stick to the extremes in comparison of the VAS to the Likert scale. On the Likert scale, children marked the middle options more frequently. Table 3 shows the correlation coefcients (r) between the response options for the seven questions. Overall, the numeric VAS correlated with the Likert scale with a coefcient of 0.82, and with the simple VAS with a

Table 2. Percentage scoring in the different categories in the Likert scale and the simple visual analogue scale (VAS).
Percentage in Likert categoriesa Item Dreams Liking teachers TV watching Current mood Playing sports Height Taking the bus
a b

Percentage in even VAS categoriesb 020 42 35 53 50 75 60 22 2140 28 27 25 19 11 17 2 4160 21 24 15 18 11 14 6 6180 4 6 5 7 1 5 13 81100 5 8 2 6 2 4 57 Wd <0.001 <0.001 <0.001 0.005 <0.001 <0.001 0.65

Percentage in uneven VAS categoriesc 010 36 28 43 35 62 47 21 1136 23 23 29 31 22 28 4 3763 32 36 21 22 13 16 7 6490 5 8 6 7 2 7 23 91100 4 6 1 5 1 2 46 Wd 0.001 0.04 <0.001 0.58 0.001 0.009 0.04

1 17 16 31 23 53 30 17

2 35 32 36 48 31 46 3

3 43 40 18 19 13 13 5

4 3 7 15 8 3 8 28

5 3 6 0 3 2 3 47

15: answering options for the Likert scale, 1 = most positive answer, 5 = most negative answer. 020, etc.: even distribution of the 10-cm line of the simple VAS. c 010 etc.: uneven distribution of the 10-cm line of the simple VAS. d Wilcoxon statistics p-value between children scoring in categories of the Likert compared with those scoring in the evenly distributed VAS, or in the unevenly distributed VAS, respectively.

834

H van Laerhoven et al.

ACTA PDIATR 93 (2004)

coefcient of 0.80. The simple VAS and the Likert scale correlated with a coefcient of 0.76.

Discussion
The results of this study show that both younger and older children prefer the Likert scale and nd it easiest to complete. The Likert scale and the VAS correlate strongly and can therefore be considered comparable with respect to reliability in children. When questionnaires are designed, feasibility, validity and reliability are important. Patient preference and ease of completion are good indicators of an instruments feasibility. In our study, all children, regardless of age, preferred the Likert scale and found it easiest to complete. This result does not support our hypothesis of age-related response patterns. The amount of missed questions is another important indicator of an instruments feasibility. Three percent of all questions in the simple VAS were missed. This was a higher percentage than the number of missed questions in the Likert type of response option (0.5%). This indicates that some children found it more difcult to answer questions in the simple VAS. However, there are some limitations to this study that need to be discussed. First of all, the number of participating children was relatively small, making it difcult to draw rm conclusions. Second, because the selection of the convenience sample consisted of children with various diseases and disease severities, our ndings can only be applied to ill children, not to healthy ones. Third, the validity of the individual response options could not be tested, as we did not have a true answer to the seven questions asked. Fourth, only relative reliability could be computed, as test retest procedures could not be performed. Testretest procedures should be performed under identical circumstances. As the rst assessment took place at an outpatient clinic, we did not consider it ethically justiable to ask the children to return to the clinic after 2 wk for the purpose of this study only. As our hospital serves as a secondary care centre for a specic area in Amsterdam, our sample consisted of a high proportion of immigrant children. There was a difference between immigrant and native children in

preference and difculty scores for the different response options. Immigrant children rated the simple VAS higher, and they also needed more time to complete the questionnaires. Perhaps immigrant childrens poorer verbal skills could explain their preference for the simple VAS. Finally, it may be discussed whether correlation was inuenced by the lack of time between completion of the different questionnaire pages, allowing children to peek at their previous answers, whereby correlation coefcients were articially increased. Other investigators introduced an interval between the different response options in earlier studies (2, 810). However, due to the moral objection of asking children to return to the clinic, this time interval was not feasible within the context of this study. However, our study investigator was present when the children lled out the questionnaire pages and prevented them from peeking. Furthermore, Downie et al. showed that correlation is maintained whether completion of the scales is separated in time or not (3). We therefore consider the high correlation coefcients found in our study as true, high relative reliability between response options. Pantel and Lewis identied a position bias of childrens responses in questionnaires, where they tended to choose the rst answer among response options (18). This effect was strongest in younger children. Other authors found that when using the simple VAS, children and the elderly stick to the anchors without using the full scale range (3, 6, 11). On the other hand, the VAS and Likert scales often encounter end-aversion bias among adults, where people tend not to mark the two extreme ends of the scale (19). We found that children stick to the rst answering category on the simple VAS, which remained apparent after the anchors of the VAS had been rearranged into smaller categories. Thus, the potential advantage of an increased answering precision of the simple VAS is lost. On the Likert scale, more children used options two and three (a little bit, sometimes), suggesting end-aversion bias. This problem might be overcome by increasing the number of answering categories from ve to seven or even nine (19). In conclusion, the relatively small number of patients makes it difcult to arrive at any nal conclusions. However, our results indicate a clear preference of

Table 3. Correlations between the response options for the seven questions.
Dreams A A B C 1 B 0.67 1 C 0.71 0.73 1 A 1 Teachers in school B 0.82 1 C 0.86 0.90 1 A 1 Frequency of watching TV B 0.71 1 C 0.82 0.77 1 Current mood A 1 B 0.75 1 C 0.85 0.80 1 A 1 Sports B 0.73 1 C 0.79 0.80 1 A 1 Height B 0.80 1 C 0.79 0.75 1 A 1 Frequency of going by bus B 0.84 1 C 0.89 0.87 1

Spearman rank order correlation coefcients are presented. A = Likert scale, B = simple VAS, C = numeric VAS.

ACTA PDIATR 93 (2004)

Response options in childrens questionnaires

835

children for the Likert scale. The Likert scale was found easiest to complete. Correlation coefcients between the Likert scale, simple VAS and numeric VAS were high, suggesting comparable reliability of the response options. The simple VAS, which we expected to be popular among young children, was not preferred by this group. Therefore, we recommend the use of Likert scale response options in questionnaires for children, although research into larger and more selected samples is needed.
Acknowledgements.HJ van der Zaag-Loonen was funded by the Dutch Digestive Diseases Foundation (Maag-Lever-Darm Stichting), Nieuwegein, The Netherlands, grant number WS 99-23. We thank Julia Loonen for editing the nal manuscript.

9. 10. 11.

12.

13.

References
1. Chambers CT, Johnston C. Developmental Differences in Childrens Use of Rating Scales. J Pediatr Psychol 2002; 27: 2736 2. Carlsson AM. Assessment of chronic pain. I. Aspects of the reliability and validity of the Visual Analogue Scale. Pain 1983; 16: 87101 3. Downie WW, Leatham PA, Rhind VM, Wright V, Branco JA, Anderson JA. Studies with pain rating scales. Ann Rheumatic Dis 1978; 37: 37881 4. Duncan GH, Bushnell MC, Lavigna GJ. Comparison of verbal and visual analogue scales for measuring the intensity and unpleasantness of experimental pain. Pain 1989; 37: 295303 5. Huskisson EC. Measurement of pain. Lancet 1974; 2: 112731 6. Kremer E, Atkinson J, Ignelzi RJ. Measurement of pain: patient preference does not confound pain measurement. Pain 1981; 10: 2419 7. Ferraz MB, Quaresma MR, Aquino LRL, Atra E, Tugwell P, Goldsmith CH. Reliability of pain scales in the assessment of literate and illiterate patient with rheumatoid arthritis. J Rheumatol 1990; 17: 10224 8. Guyatt GH, Townsend M, Berman LB, Keller JL. A comparison

14.

15. 16. 17. 18. 19.

of Likert and visual analogue scales for measuring change in function. J Chron Dis 1987; 40: 112933 Jaeschke R, Singer J, Guyatt GH. A comparison of seven-point and visual analogue scales: a randomised controlled trial. Contr Clinical Trials 1990; 11: 4351 Paul-Dauphin A, Guillemin F, Virion JM, Briancon S. Bias and precision in Visual Analogue Scales: a randomised controlled trial. Am J Epidemiol 1999; 150: 111727 Apajasalo M, Rautonen J, Holmberg C, Sinkkonen J, Aalberg V, Pihko H, et all.. Quality of life in pre-adolescence: a 17-dimensional health-related measure (17D). Qual Life Res 1996; 5: 5328 Landgraf JM, Abetz LN. Measuring health outcomes in pediatric populations: issues in psychometrics and application. In: Quality of life assessments and pharmacoeconomics in clinical trials. 2nd ed. New York: Raven Press; 1996: 793802 Rebok G, Riley A, Forrest C, Stareld B, Green B, Robertson J, Tambor E. Elementery school-aged schildrens reports of their health: a cognitive interviewing study. Qual Life Res 2001; 10: 5970 Juniper EF, Guyatt GH, Feeny DH, Grifth LE, Ferrie PJ. Minimum skills required by children to complete health-related quality of life instruments for asthma: comparison of measurement properties. Eur Resp J 1997; 10: 228594 Marsh HW. Negative item bias in rating scales for preadolescent children: a cognitive-developmental phenomenon. Develop Psychol 1986; 22: 3749 Ross DJ, Ross SA. Childhood pain: the school-aged childs viewpoint. Pain 1984; 20: 17991 Abu-Saad HH, Kroonen E, Halfens R. On the development of a multidimensional Dutch pain assessment tool for children. Pain 1990; 43: 24956 Pantell RH, Lewis CC. Measuring the impact of medical care on children. J Chron Dis 1987; 40 Suppl 1: 99108S Streiner DL, Norman GR. Health measurement scales: a practical guide to their development and use. 2nd ed. New York: Oxford University Press; 1995

Received Nov. 15, 2002; revisions received May 26, 2003 and Dec. 29, 2003; accepted Jan. 13, 2004