 Research
 Open Access
 Published:
Reliability of 95% confidence interval revealed by expected qualityoflife scores: an example of nasopharyngeal carcinoma patients after radiotherapy using EORTC QLQC 30
Health and Quality of Life Outcomes volume 8, Article number: 68 (2010)
Abstract
Background
Many researchers use observed questionnaire scores to evaluate score reliability and to make conclusions and inferences regarding qualityoflife outcomes. The amount of false alarms from medical diagnoses that would be avoided if observed scores were substituted with expected scores is interesting, and understanding these differences is important for the care of cancer patients. Using expected scores to estimate the reliability of 95% confidence intervals (CIs) is rarely reported in published papers. We investigated the reliability of patient responses to a qualityoflife questionnaire and made recommendations for future studies of the quality of life of patients.
Methods
A total of 115 patients completed the EORTC core questionnaire QLQC30 (version 3) after radiotherapy. The observed response scores, assumed to be onedimensional, were summed and transformed into expected scores using the Rasch rating scale model with WINSTEPS software. A series of simulations was performed using a unified bootstrap procedure after manipulating scenarios with different questionnaire lengths and patient numbers to estimate the reliability at 95% confidence intervals. Skewness analyses of the 95% CIs were compared to detect different effects between groups according to the two data sets of observed and expected response scores.
Results
We found that (1) it is necessary to report CIs for reliability and skewness coefficients in papers; (2) data derived from expected response scores are preferable to making inferences; and (3) visual representations displaying the 95% CIs of skewness values applied to itembyitem analyses can provide a useful interpretation of qualityoflife outcomes.
Conclusion
Reliability coefficients can be reported with 95% CIs by statistical software to evaluate the internal consistency of respondent scores on questionnaire items. The SPSS syntax procedures for estimating the reliability of the 95% CI, expected score generation and visual skewness analyses are demonstrated in this study. We recommend that effect sizes such as a 95% CI be reported along with p values reporting significant differences in qualityoflife studies.
Background
Cronbach's α coefficient (hereinafter referred to as Alpha [1]) is widely used as an index of scoring reliability and is often reported in social and behavioral studies [2, 3]. However, very few authors report the confidence intervals (CIs) of Alpha in their papers, although this has been suggested by many researchers [4–7].
Conception of research questions
The American Psychological Association Task Force on Statistical Inference suggests, "ways provide some effectsize estimate when reporting a p value" [4]. The task force also noted that "interpreting the size of observed effects requires an assessment of the reliability of the scores" because score unreliability weakens effect size [5]. Confidence intervals around Alpha are advocated for both absolute and relative decisions made by researchers [6, 7]. Alexander et al., in their critical review of research methods regarding quality improvement research, also stated that 74.1% of the studied articles discussed effect size [8]. These reports underscore the need for research on the disclosure of Alpha and 95% CIs in qualityoflife studies.
Most research on quality of life uses observed scores obtained from surveys and questionnaires to make conclusions and inferences regarding qualityoflife outcomes. False responses on patient questionnaires can be caused by aberrant answers, guessing, inattentiveness (which may result from fatigue), misunderstanding or response misfit, according to item response theory [9, 10]. These issues can result in false alarms (i.e., false positive and false negative errors) coming from medical diagnoses and leading to make vain efforts in actions taking. It is of interest to discuss the possibility of replacing observed response scores with expected ones to make decisions and inferences by applying probability theory to studies of qualityoflife outcomes.
Objectives
This article is organized as follows. First, a set of EORTC QLQC30 component items developed to assess the quality of life of cancer patients [11] was given to 115 patients who were diagnosed with nasopharyngeal carcinoma. Second, we present a brief introduction of the Rasch [12] model, which we performed to determine expected response scores by the probability function of item difficulty and person ability. Third, Alphas for 95% CIs were generated by a SPSS syntax procedures for reliability 95% CI estimation referring to both observed and expected response scores on EORTC QLQC30 by manipulating different person numbers. Fourth, graphical representations of skewness analyses with 95% CIs were applied to justify differences between cancer stages across items, by which the misfitted items on the EORTC QLQC30 were examined in further detail with both observed and expected response scores.
Methods
Research data resource
A total of 115 patients newly diagnosed with nasopharyngeal carcinoma were recruited to answer the first 28 questions on the Taiwanese version of the EORTC QLQC30 questionnaire, which requires answers on a scale of 1 to 4 (1 = Not at all, 2 = A little, 3 = Quite a bit, and 4 = Very much). The two composite questions requiring answers on a 17 scale were excluded from the study ("How would you rate your overall health during the past week?" and "How would you rate your overall quality of life during the past week?"). All examinees were treated in one of three Taiwanese medical centers and had a followup evaluation a minimum of three months later without complete remission. Accordingly, the 115person × 28item observedresponse matrix was created. This study was approved by the institutional review boards of the hospitals.
Data analysis using the Rasch model
We assumed that the 28item EORTC QLQC30 fits Rasch model's specification and forms a unidimensional construct. If any item could not fit the Rasch model expectation, patients were deemed to have exhibited unexpected behaviors, such as aberrant responses, guessing or inattentiveness, which may have led to responses outside of the model's expectations (as we assumed all 28 items constructed a onedimensional latent trait).
When the data fit the model's expectations, the infit and outfit mean square error (MNSQ) statistics had an expected value of unity on the items. The values of the MNSQ statistics show the amount of distortion of the measurement system. Values less than unity (or overfit) showed that the items were too predictable (i.e., there was redundancy). Values greater than unity (or underfit) indicated unpredictability (i.e., noise). The MNSQ statistics were chisquare statistics divided by their degrees of freedom. Items with infit MNSQ beyond the range of 0.51.5 were usually regarded as misfitting or poorfitting [13, 14].
Excel module for data simulation
For our approach to succeed, an ExcelVBA(visual basic for application) module according to the formulas of Cronbach's α (see Formula 1) and skewness (see Formula 2) was programmed.
Cronbach's α is defined in Formula 1, where N is the number of items, {\sigma}_{x}^{2} is the variance of the observed total test scores for the sample of examinees and {\sigma}_{Yi}^{2} is the variance of item i for the sample of examinees.
Formula 2 shows the equation for skewness:
where n is the number of responses, x _{ i }is the value of individual responses and \overline{x} and s denote the mean and standard deviation of the total responses, respectively.
In Formula 3, based on the Rasch rating scale model [15] shown in Additional file 1, the probability of a person n in a specific category j of item i is yielded by person ability θ and item difficulty δ on an mpoint scaled questionnaire.
The expected response scores would be generated by Rasch model's probability function of item difficulties and person measures (see Additional file 1). We anticipate that the person expected scores could produce item MNSQ statistics between 0.5 and 1.5 when item difficulties are anchored according to the original 115 (person) × 28 (item) observed response data set.
The authorprogrammed A ExcelVBA module for computing test Cronbach's α, item/person skewness coefficient and person expected response score generation is shown (in Additional file 2) according to the item difficulties (Table 1) yielded by WINSTEPS [16] using the Rasch rating scale model [15] from the original 115 (person) × 28 (item) observed matrix.
Procedures for data simulation
1. Alpha 95% CIs by both observed and expected scores
The present study built on the data that were collected from the 115 patients answering the aforementioned 28 items on the EORTC QLQC30. As mentioned above, we investigated 95% CIs from the perspective of Alpha by examining different person numbers, such as 115, double(230) and trible(445) with a SPSS syntax procedures (see Additional file 3) referring to both observed and expected scores on EORTC QLQC30 to examine any difference between scenarios.
2. Generating expected response data sets by simulation
The random response, for instance with a dichotomous scale, less than the probability {P}_{ni}=\frac{\mathrm{exp}({\theta}_{n}{\delta}_{i})}{1+\mathrm{exp}({\theta}_{n}{\delta}_{i})}, an item response of 1 would be assigned for that examinee on the item. On the other hand, had the random number been greater than the response probability, an item response of 0 would be recorded. This is a standard item response generation method as used by previously published papers [17–19]. We can easily extend it to a polytomylike EORTC QLQC30 to generate the expected item response data. The Rasch rating scale model [15] probability generation program is shown in Additional file 1 and was applied to yield a data set of expected response scores constructed by the 14 scaling scheme.
3. Itembyitem skewness analysis by a bootstrapping procedure
To compare the quality of life on the 28item EORTC QLQC30 questionnaire, we applied itembyitem comparisons [20, 21] of the patient attitudes according to cancer stage. For this purpose, skewness analyses were performed by a bootstrapping method [22–24], sampling with replacement of patient responses at each item, each time selecting 500 responses with 100 replications in ascending order to filter out the median skewness and its 95% CI.
The correlation coefficients between item difficulties (in logit (log odds) units calibrated by WINSTEPS), observed and expected response summation scores across items and the skewness yielded by both observed and expected responses on items are reported. We will examine whether the skewness of expected responses can earn a higher correlation to other counterparts than the skewness of observed scores. A graphical representation of the comparison of patient attitudes toward quality of life across the entirety of the questionnaire will be presented.
Statistical analyses
Statistical analyses and plotting were performed by SPSS software for Windows (Version 15, SPSS, Chicago, IL).
Results
Item difficulties of EORTC QLQC30
The most difficult item was question 5 (Do you need help with eating, dressing, washing yourself or using the toilet?), as shown in Table 1. The easiest question was 18 (Were you tired?). Expected response scores, as anticipated, presented an acceptable infit MNSQ more often than observed scores, which displayed two items outside of the range 0.51.5. From the observed score perspective, question 5 had an infit MNSQ of 1.6, and question 28 had an infit MNSQ of 1.74, indicating some aberrant responses on those two questions that negatively affected the inferences and conclusions that were made.
Alpha values for different person numbers
In Table 2 we can see that expected scores earn higher Alpha and its 95% CIs across person numbers than the observed responses. The more person number can produce the narrower 95% CI band.
Skewness analysis with 95% CI on misfitting items
As shown in Table 1, questions 5 and 28 exhibited greater infit MNSQ errors, indicating some nonrandomized, or, say, unexpected aberrant responses on these two items. The skewness and 95% CI of these two misfitted items, calculated from the two data sets of observed and expected scores, are shown on both sides of Figure 1. Significant differences were seen in stage I by the 95% CI far from others of stage II, III and VI on item 5, for which skewness only in stage II displayed a lower value (3.18) by observed scores than by expected scores, (4.32). On item 28, observed raw scores showed a significant difference between stages I and II, but the expected scores did not. When using Rasch measures to generate expected scores for prediction, slightly better results might be obtained when the measure estimates are based on adjusted data. A productive adjustment makes observations in extreme categories slightly less extreme[16].
Overall skewness analysis with 95% CI across items
In Figure 2 we can see that there were 6 items on stages showing significant differences in qualityoflife attitude by the expected response scores, but 11 items on stages gave significant differences by the observed response scores, indicating different inferences would have been made if we had chosen different data sources (i.e., observed and expected scores in Figure 2).
As shown in Table 3, the skewness for expected response scores had a higher correlation coefficient (≥ 0.90) than the skewness with observed scores (some ≤ 0.90), as compared to other counterparts, such as Rasch transformed item difficulties (which were derived from original observed response scores), observed and expected summation response scores. This indicates that the expected scores would be more meaningful than the observed scores for making inferences.
Discussion
The reliability of a test in a given population is defined as the ratio of true score variance to observed score variance [25]. In empirical research, true scores cannot be directly determined. One of the most frequently used methods of estimating internal consistency reliability is Cronbach's coefficient α, as shown in Formula 1, which is frequently cited (at least 5,590 times, including approximately 325 social science citations, per year) [2]. Our data indicate that CIs for Cronbach's α should be more widely calculated and reported for published data. The SPSS syntax procedure of intervals for reliability coefficients is included in Additional file 3, although it can be estimated in various ways [7]. We hope that in the future, authors will report Cronbach's α routinely and note what estimation methods they use. This would help clarify their score reliabilities and how these are affected by sampling error variance.
Findings
We utilized the Rasch model, using the term "ability" for a latent trait that underlies the questionnaire responses of the participants, to determine the probability of each examinee answering a given question accurately [26]. The expected response scores could then be derived from the original observed data matrix, which we used to obtain a more accurate Alpha and its intervals. Furthermore, the graphical representation of itembyitem skewness analyses using expected response scores was plotted to make more appropriate inferences and conclusions of difference detection among groups of interest, particularly on misfit responses.
Strengths of the study
Most data analysis of qualityoflife questionnaires is based on classic test theory (CTT). In recent years, CTT has been gradually replaced by item response theory (IRT) [10, 14, 27, 28]. Whereas CTT concerns the accuracy of observed scores, IRT concerns the accuracy of ability scores [29]. In this study, we applied IRTbased Rasch analysis to generate expected response scores and confidence intervals, approaching a more accurate understanding of our patient population and enabling proper treatment decisions.
From Figure 1 and 2 we can see that the 95%CIs for an onesamnple case far from the skewness value of zero is deemed that a statistically significant difference in an attitude toward positive (with a negatively skewed distribution if higher scores represent positive) or toward negative (with a positively skewed distribution if lower scores represent negative) is found (p < .05). However, for the twosample case it should be cautious about interpreting the graphical reports when examining the attitude difference between groups by nonoverlapping 95% confidence intervals (CIs). According to the criterion of 95% CIs derived from two independent samples(e.g., A and B), there is a distance beyond 2*SE_A + 2*SE_B between those two means with more than 95% confidence to regard as a significant difference emerged between two groups. That is said that (1) bootstrapping approach is especially suited to data following a nonnormal distribution; (2) the strict criterion of 95% CIs derived from two independent samples shows that MB  MA > 2 × SE_A + 2 × SE_B > 2 × sqrt(SE_A^2 + SE_B^2) which holds a high confidence to regard as a significant difference emerged in two groups by checking nonoverlapping 95% confidence intervals (CIs).
Limitations of the study
We simulated data from 115 patients newly diagnosed with nasopharyngeal carcinoma who responded to the 28 questions on the Taiwanese version of the EORTC QLQC30 questionnaire that requires answers on a scale of 1 to 4. Some bias might have occurred as a result of the Taiwanese translation of the questionnaire. If so, the itemcalibrated difficulties, including step threshold difficulties, would be different among patients from different countries, which may lead to a variety of expected response scores by the Rasch model. The Alpha and 95% CIs shown in Table 2, according to sample size, would also be different in relation to different types of response outliers [9], different test forms, different numbers of questions, misspecifications and confounding independent variables in a single reliability generalization analysis [28].
The Cronbach's alpha has been frequently reported in CTT for a long time though Rasch person separation reliability was reported better than Cronbach's alpha as a test indication in recent years [28]. We do not decline the use of WINSTEPS command by SIFILE = data file name to simulate a data file equivalent to the raw data to obtain Rasch person separation reliability 95% CIs when demonstrating the use of Cronbach's alpha 95% CIs in Additional file 3.
In this study we assumed that the 28item EORTC QLQC30 fits Rasch model's specification and forms a unidimensional construct. However, to our knowledge till now, it has not been either previously tested or reported using a Rasch analysis rather than sample and item dependent classical test models. For the reason of space limitation for this manuscript, we have not explored the dimensionality and invariance of the EORTC QLQC30 like the General Health Questionnaire (GHQ)12 did [30] in detail through the Rasch model in this paper. A brief report regarding dimension and invariance of items has been enclosed in Additional file 4.
Applications
There are concerns that the expected response scores derived from observed raw scores may also yield unreliable results. The frequent counts of total items (similar to the item difficulties in this study) endorsed by examinees would not be overly distorted if we assume that the errors made by them were randomized across the questions. The specific misfit items observed by a particular group (e.g., stage II cancer patients on item 28 in Figure 2) could be well adjusted by expected response scores generated by the Rasch model's probability theory. In addition, the shortcoming of missing data handled in CTT could be overcome through the derived measures estimated by Rasch modeling.
The SPSS syntax procedures in Additional file 3, which has been revised referring to the previously published article [7], can help authors easily report Alpha intervals in studies involving qualityoflife investigation and surveillance. Furthermore, the skewness coefficient and its 95% CI, referring to expected response scores, can be used to verify whether the assessed attitude is neutral (approaching a normal distribution), in absolute agreement (with a negatively skewed distribution if higher scores represent agreement) or in disagreement (with a positively skewed distribution if lower scores represent disagreement). Use of this coefficient merits further study in other kinds of qualityoflife surveys. One major obstacle is that users may need some training to interpret the visual representation of the skewness analysis.
Conclusion
Intervals for reliability coefficients can be estimated in various ways. We demonstrated bootstrap and SPSS syntax procedures in this study. There is no particular reliability coefficient or CI estimation method that authors are expected to universally invoke [31, 32]. After all, each of the factors in a qualityoflife assessment was different in our study; however, they did not independently vary. We suggest that authors routinely report CIs in qualityoflife publications, to clarify the statistical estimates of the effect of sampling error variance on score reliability. These requirements may also facilitate the understanding that scores in a questionnaire are not completely associated with invariant reliability. Furthermore, using expected response scores is encouraged. Graphical skewness representation with 95% CI is also encouraged to depict patient quality of life in future studies.
Abbreviations
 CTT:

classic test theory
 Alpha:

Cronbach's α
 IRT:

item response theory
 CI:

confidence interval
 MNSQ:

mean square error.
References
 1.
Cronbach LJ: Coefficient alpha and the internal structure of tests. Psychometrika 1951, 16: 297–334. 10.1007/BF02310555
 2.
Cronbach LJ, Shavelson RJ: My current thoughts on coefficient alpha and successor procedures. Educational and Psychological Measurement 2004, 64: 391–418. 10.1177/0013164404266386
 3.
Zumbo BD, Rupp AA: Responsible modelling of measurement data for appropriate inferences: Important advances in reliability and validity theory. In The Sage handbook of quantitative methodology for the social sciences. Edited by: Kaplan D. Thousand Oaks CA: Sage; 73–92.
 4.
Wilkinson L, APA: Task Force on Statistical Inference: Statistical methods in psychology journals: Guidelines and explanations. American Psychologist 1999, 54: 594–604. 10.1037/0003066X.54.8.594
 5.
Cook C, Heath F, Thompson RL, Thompson B: Score reliability in web or Internetbased surveys: unnumbered graphic rating scales versus LikertType scales. Educational and Psychological Measurement 2001, 61: 697. 10.1177/00131640121971356
 6.
Burdick RK, Graybill FA: Confidence intervals for variance components. New York: Marcel Dekker; 1992.
 7.
Fan X, Thompson B: Confidence intervals about score reliability coefficients please: EPM guidelines editorial. Educational and Psychological Measurement 2001,61(4):517–531.
 8.
Alexander JA, Hearld LR: Review: what can we learn from quality improvement research? a critical review of research methods. Med Care Res Rev 2009, 66: 235–271. 10.1177/1077558708330424
 9.
Liu Y, Wu AD, Zumbo BD: The impact of outliers on Cronbach's Coefficient Alpha estimate of reliability: ordinal/rating scale otem Responses. Educational and Psychological Measurement 2007,67(4):620–634. 10.1177/0013164406296976
 10.
Chien TW, Wang WC, Lin SB, Lin CY, Guo HR, Su SB: KIDMAP a Web based system for gathering patients' feedback on their doctors. BMC Med Res Methodol 2009,9(1):38. 10.1186/14712288938
 11.
Bjordal K, de Graeff A, Fayers PM, et al.: A 12 country field study of the EORTC QLQC30(version 3.0) and the head and neck cancer specific module(EORTC QLQH&HN35) in Head and Neck patients. Eur J Cancer 2000, 36: 1796–1807. 10.1016/S09598049(00)001866
 12.
Rasch G: Probabilistic models for some Intelligence and attainment tests. Chicago: University of Chicago Press; 1960.
 13.
Wang WC, Yao G, Tsai YJ, Wang JD, Hsieh CL: Validating, Improving Reliability and Estimating Correlation of the Four Subscales in the WHOQOLBREF using Multidimensional Rasch Analysis. Quality of Life Research 2006,15(4):607–620. (14) 10.1007/s1113600543657
 14.
Wright BD, Masters GN: Rating scale analysis Rasch measurement. Chicago: MESA Press; 1982.
 15.
Andrich D: A rating scale formulation for ordered response categories. Psychometrika 1978, 43: 561–73. 10.1007/BF02293814
 16.
Linacre JM: WINSTEPS [computer program]. Chicago IL 2009. [http://www.winsteps.com]
 17.
Kieffer KM, Reese RJ: A reliabilty generalization study of the ceriatric scale. Educational and Psychological Measurement 2002,62(6):969–994. 10.1177/0013164402238085
 18.
Harwell M, Stone CA, Hsu TC, Kirisci L: Monte Carlo studies in item response theory. Applied Psychological Measurement 1996, 20: 101–125. 10.1177/014662169602000201
 19.
Macdonald P, Paunonen SV: A monte carlo comparison of item and person statistics based on item response theory versus classical test theory. Educational and Psychological Measurement 2002, 62: 921–943. 10.1177/0013164402238082
 20.
Chien TW, Wang WC, Wang HY, Lin HJ: Online assessment of patients' views on hospital performances using Rasch model's KIDMAP diagram. BMC Health Serv Res 2009, 9: 135. 10.1186/147269639135
 21.
Chien TW, Wu HM, Wang WC, Castillo RV, Chou W: Reduction in patient burdens with graphical computerized adaptive testing on the ADL scale: tool development and simulation. Health Qual Life Outcomes 2009, 7: 39. 10.1186/14777525739
 22.
Chernick MR: Bootstrap Methods A practitioner's guide. Wiley Series in Probability and Statistics 1999.
 23.
Efron B: Bootstrap Methods: Another Look at the Jackknife. The Annals of Statistics 1979,7(1):1–26. 10.1214/aos/1176344552
 24.
Efron B, Tibshirani R: An Introduction to the Bootstrap. Chapman & Hall/CRC; 1993.
 25.
Lord FM, Novick R: Statistical theories of mental tests scores. Reading MA: AddisonWesley; 1968:57.
 26.
Lord FM: Applications of item response theory to practical testing problems. Hillsdale NJ: Lawrence Erlbaum; 1980.
 27.
Wright BD, Mok M: Understanding Rasch Measurement Rasch Models Overview . J Appl Meas 2000, 1: 83–106.
 28.
Schumacker RE, Smith EV Jr: Reliability: a Rasch perspective. Educational and Psychological Measurement 2007,67(3):394–409. 10.1177/0013164406294776
 29.
Dimitrov DM: Reliability: Arguments for multiple perspectives and potential problems with generalization across studies. Educational and Psychological Measurement 2002, 62: 783–801. 10.1177/001316402236878
 30.
Smith AB, Fallowfield LJ, Stark DP, Velikova G, Jenkins V: A Rasch and confirmatory factor analysis of the General Health Questionnaire (GHQ)12. Health and Quality of Life Outcomes 2010, 8: 45. 10.1186/14777525845
 31.
Thompson B: Guidelines for authors. Educational and Psychological Measurement 1994, 54: 837–847.
 32.
Kirk R: Practical significance: A concept whose time has come. Educational and Psychological Measurement 1996, 56: 746–759. 10.1177/0013164496056005002
Acknowledgements
This study was supported by Grant 96 cmkmu17 from the Chi Mei Medical Center, Taiwan.
Author information
Affiliations
Corresponding author
Additional information
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
WC, SJ and LF collected all data, generated the database, designed and performed the statistical analysis and wrote the manuscript. TW, SJ,WW and LF contributed to the development of the study design and advised on the performance of the statistical analysis. The analysis and results were discussed by all authors together. TW, WW and WP contributed to the Excel programming, helped to interpret the results and helped to draft the manuscript. All authors read and approved the final manuscript.
ShunJin Lin and Agnes LF Chan contributed equally to this work.
Electronic supplementary material
Expected scores obtained by the Rasch model's probability theory
Additional file 1: . ExcelVBA program for randomly generating Rasch model's expected scores. (DOC 28 KB)
Authormade A ExcelVBA module computing test Cronbach's
Additional file 2: α , item/person skewness coefficient and person expected response score. ExcelVBA routine of a calculation demonstration for skewness and its plotting. (XLS 394 KB)
SPSS syntax procedures for estimation of the reliability of the 95% CI
Additional file 3: . A example of SPSS syntax procedures for Cronbach α and its 95% CI. (DOC 24 KB)
The dimensionality and invariance of the EORTC QLQC30 explored briefly through the Rasch model
Additional file 4: . A brief analysis for the dimensionality of the EORTC QLQC30. (DOC 66 KB)
Authors’ original submitted files for images
Below are the links to the authors’ original submitted files for images.
Rights and permissions
Open Access This article is published under license to BioMed Central Ltd. This is an Open Access article is distributed under the terms of the Creative Commons Attribution License ( https://creativecommons.org/licenses/by/2.0 ), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
About this article
Cite this article
Chien, TW., Lin, SJ., Wang, WC. et al. Reliability of 95% confidence interval revealed by expected qualityoflife scores: an example of nasopharyngeal carcinoma patients after radiotherapy using EORTC QLQC 30. Health Qual Life Outcomes 8, 68 (2010). https://doi.org/10.1186/14777525868
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/14777525868
Keywords
 Item Response Theory
 Item Difficulty
 Classic Test Theory
 Response Score
 Skewness Coefficient