DOCSLIB.ORG

Automatic Classification of Disaster-Related Tweets

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Automatic Classification of Disaster-Related Tweets International conference on Innovative Engineering Technologies (ICIET’2014) Dec. 28-29, 2014 Bangkok (Thailand) Automatic Classification of Disaster-Related Tweets Beverly Estephany Parilla-Ferrer, Proceso L. Fernandez Jr., PhD, and Jaime T. Ballena IV, PhD Abstract— The social networking site Twitter has become one of of its growing ubiquity, communications rapidity, and cross- the quickest sources of news and other information. Twitter platform accessibility [1]. The interactions on social media information feeds known as tweets, are voluntarily sent by registered being highly distributed, decentralised and occurring in real users and reach even non-registered users, sometimes ahead of time, provide the necessary breadth and immediacy of traditional sources of mass news. In this study, we develop some machine learning models that can automatically detect informative information required in times of emergencies [2]. disaster-related tweets. Twitter is one microblogging service that allows its A dataset of tweets, collected during the Habagat flooding of subscribers to broadcast short messages, called tweets, of up Metro Manila in 2012, was used in building the classifier models. A to 140 characters. These tweets are used to share relevant random subset of this dataset was manually labeled as either information and report news [3]. In emergency situations, informative or uninformative to produce the ground truth. Two tweets provide either first-person observations or bring machine learning algorithms, Naive Bayes and Support Vector Machine (SVM), were used to build models for the automatic relevant knowledge from external sources [1]. Twitter is classification of the tweets, and these models were evaluated across becoming a valuable tool in disaster and emergency situations, the metrics of accuracy, precision, recall, area under curve and F- as there is increasing evidence that it is not just a social measure. Experimental results show that the model generated from network, it is also a news service [4]. Relevant tweets shared SVM has significantly better results compared to that of the Naive by users is a vital source of information and is useful in Bayes. understanding and visualizing the situation of affected parties. This study also revealed that uninformative tweets outnumbered informative tweets, suggesting that the subscribers used Twitter to This medium is seen as a place for ―harvesting‖ information broadcast more of tweets that express their subjective messages and during a crisis event to determine what is happening on the emotions regarding the Habagat event. However, the informative ground [5]. The growing use of social media during crises tweets were more likely to be retweeted than uninformative tweets, offers new information sources from which the right indicating that subscribers retweet messages they deem informative authorities can enhance emergency situation awareness which and useful for public awareness. These insights, together with the is significantly recognized as a critical part of making built classifier models, can help in the development of a system that can sift through the voluminous Twitter data and in real-time detect successful and effective decisions for emergency response [6]. informative disaster-related tweets so that appropriate action may be Tweets highly vary in terms of subject and content and the done promptly. influx of tweets particularly in the event of a disaster may be overwhelming. It consists of socio-behaviors that include Keywords— Disaster, Machine learning, Text mining, Tweets intensified information search and information contagion [7]. Microblogging offers ways to retrieve, produce and spread I. INTRODUCTION information; the nature of that sharing has a lifecycle of URING disasters and emergencies, microblogs, have information production and consumption that is rapid and D been used by people whether from the private or public repetitive [1]. Since these varied tweets are rapidly sector, local or international community, as a medium to broadcasted, it is imperative to automatically classify the broadcast their messages. This social medium is being tweets in order to extract needed information. The availability considered as a means for emergency communications and accessibility of disaster-relevant information can because contribute to an effective and efficient disaster response mechanism, which eventually can alleviate damages or loss of life and property during a disaster or crisis. Beverly Estephany Parilla-Ferrer is with the Department of Information Disaster-related tweets are one of the many subjects of text Technology and Computer Science, School of Information and Computing mining researches nowadays. Specifically on the area of Science Saint Louis University, Baguio City, Philippines(+637409205884351; e-mail:[email protected]). classifying and extracting information from disaster-related Proceso L. Fernandez Jr., is with the Department of Information Systems tweets, Caragea et al conducted a study to classify the set of and Computer Science, School of Science and Engineering, Ateneo de Manila tweets collected during the Haiti earthquake for the University, Quezon City, Philippines (e-mail:[email protected]). Jaime T. Ballena IV is with the Math Department, School of Information emergency response sector [8]. The authors compared and Computing Science Saint Louis University, Baguio City Philippines (e- different feature representations for training SVM classifiers mail: [email protected]). http://dx.doi.org/10.15242/IIE.E1214072 62 International conference on Innovative Engineering Technologies (ICIET’2014) Dec. 28-29, 2014 Bangkok (Thailand) to classify tweets. Imran et al extracted information using and classifies tweets into user-defined categories. The disaster-related message ontology to classify tweets using the experimental results returned 75% classification accuracy Joplin dataset [9]. Multi-level and multi-label classification based on AUC metric. Prasetyo et al used SVM algorithm to using Naive Bayes classifier in Weka was used in the study. classify software related microblogs as relevant or irrelevant Another paper of Imran et al focused on the classification and to engineering software systems with text from URLs and extraction of disaster-relevant information from the Joplin and microblogs as features [15]. Training and testing was Sandy dataset using conditional random fields for training performed using 10-fold cross validation and the model [10]. revealed a significantly good performance based on accuracy, These studies have clearly presented that disaster–relevant precision, recall and F-measure. information can be classified and can provide information that On the area of comparing machine learning algorithms for can augment people's awareness on incidents. However, these classification using short text messages as dataset, the studies did not cover statistical analysis on the big data of following researches were conducted. Duwairi and Qarqaz tweets and the performance evaluation of machine learning compared Naive Bayes, SVM and K-nearest Neighbor as algorithms for the classification of tweets. Although there are implemented in Rapidminer to classify sentiments of tweets as several studies that have evaluated machine learning positive, negative or neutral using a dataset on general topics algorithms for the classification of tweets, these studies dealt such as education, sports and political news [16]. With 10-fold with sentiment or opinion analysis. In this study, we aim to cross validation, SVM returned the highest precision, while K- create a machine learning model to classify disaster-related Nearest Neighbor (KNN) with the highest recall. A study on tweets as informative or uninformative and compare the the classification of Reuters headline news as dataset, Khamar performance of two of the most common machine classifying compared SVM, K-Nearest Neighbor and other algorithms algorithms Naive Bayes and Support Vector Machine. [17]. After a process of training and testing, KNN returned a Performance evaluation is based on the validation of results higher accuracy compared to Naive Bayes and SVM. Lu across the metrics of accuracy, precision, recall, area under conducted a study to identify online messages using C4.5, curve and F-measure, with the application of statistical tools. Naive Bayes and SVM [18]. Based on experiment, SVM Furthermore, the research investigates the information that can outperformed C4.5 and Naive Bayes in terms of accuracy and be extracted from the statistics of broadcasted tweets during F-measure. Zielenski and Bugel investigated on classifying the Habagat incident which caused widespread flooding in tweets posted in 4 different languages (English and 3 Metro Manila in 2012. Mediterranean languages in Turkey, Greece and Romania) as relevant or not relevant to an earthquake event by testing a II.RELATED WORKS language-specific detection classifier with keywords that are There are several researches on text mining for synonyms or translations of the word earthquake as features in classification and prediction on various domains such as in the the classification [19]. Training and testing used different medical, business, crime investigation, e-mail detection, etc. datasets using regular expression and Naive Bayes. The The following works are focused on the classification of results showed a best performance on the official languages tweets and the comparison of classifying algorithms.
Load more

Recommended publications
  • Volume Quantification by Contrast-Enhanced Ultrasound
    Herold et al. Cardiovascular Ultrasound 2013, 11:36 http://www.cardiovascularultrasound.com/content/11/1/36 CARDIOVASCULAR ULTRASOUND RESEARCH Open Access Volume quantification by contrast-enhanced ultrasound: an in-vitro comparison with true volumes and thermodilution Ingeborg HF Herold1*, Gianna Russo2, Massimo Mischi2, Patrick Houthuizen3, Tamerlan Saidov2, Marcel van het Veer3, Hans C van Assen2 and Hendrikus HM Korsten1,2 Abstract Background: Contrast-enhanced ultrasound (CEUS) has recently been proposed as a minimally- invasive, alternative method for blood volume measurement. This study aims at comparing the accuracy of CEUS and the classical thermodilution techniques for volume assessment in an in-vitro set-up. Methods: The in-vitro set-up consisted of a variable network between an inflow and outflow tube and a roller pump. The inflow and outflow tubes were insonified with an ultrasound array transducer and a thermistor was placed in each tube. Indicator dilution curves were made by injecting indicator which consisted of an ultrasound- contrast-agent diluted in ice-cold saline. Both acoustic intensity- and thermo-dilution curves were used to calculate the indicator mean transit time between the inflow and outflow tube. The volumes were derived by multiplying the estimated mean transit time by the flow rate. We compared the volumes measured by CEUS with the true volumes of the variable network and those measured by thermodilution by Bland-Altman and intraclass-correlation analysis. Results: The measurements by CEUS and thermodilution showed a very strong correlation (rs=0.94) with a modest volume underestimation by CEUS of −40 ± 28 mL and an overestimation of 84 ± 62 mL by thermodilution compared with the true volumes.
    [Show full text]
  • On the Sampling Variance of Intraclass Correlations and Genetic Correlations
    Copyright 1998 by the Genetics Society of America On the Sampling Variance of Intraclass Correlations and Genetic Correlations Peter M. Visscher University of Edinburgh, Institute of Ecology and Resource Management, Edinburgh EH9 3JG, Scotland Manuscript received February 3, 1997 Accepted for publication March 23, 1998 ABSTRACT Widely used standard expressions for the sampling variance of intraclass correlations and genetic correlation coef®cients were reviewed for small and large sample sizes. For the sampling variance of the intraclass correlation, it was shown by simulation that the commonly used expression, derived using a ®rst-order Taylor series performs better than alternative expressions found in the literature, when the between-sire degrees of freedom were small. The expressions for the sampling variance of the genetic correlation are signi®cantly biased for small sample sizes, in particular when the population values, or their estimates, are close to zero. It was shown, both analytically and by simulation, that this is because the estimate of the sampling variance becomes very large in these cases due to very small values of the denominator of the expressions. It was concluded, therefore, that for small samples, estimates of the heritabilities and genetic correlations should not be used in the expressions for the sampling variance of the genetic correlation. It was shown analytically that in cases where the population values of the heritabili- ties are known, using the estimated heritabilities rather than their true values to estimate the genetic correlation results in a lower sampling variance for the genetic correlation. Therefore, for large samples, estimates of heritabilities, and not their true values, should be used.
    [Show full text]
  • HERITABILITY ESTIMATES from TWIN STUDIES I. Formulae of Heritability Estimates
    HERITABILITY ESTIMATES FROM TWIN STUDIES I. Formulae of Heritability Estimates K.W. KANG. J.C. CHRISTIAN, J.A. NORTON, Jr. Departments of Medical Genetics and Psychiatry, Indiana University School of Medicine, Indianapolis, Indiana, USA Over the past 50 years a large number of methods have been proposed for estimating herita­ bility from twin studies. The present paper describes the most commonly cited of these esti­ mates as a first step in evaluating their usefulness. A critical review will then follow. Family studies of human quantitative traits have, in general, three goals: first, to determine whether a given phenotypic trait is genetically influenced; second, to determine if discrete segregating genetic factors influence the trait; and third, to determine whether the trait is linked with some other genetic trait. Twin data are often used to attain the first goal by measuring the relative importance of heredity and environment on the development of quan­ titative traits. The concept of heritability originated in an attempt to describe the degree to which the differences actually observed between individuals arose from differences in genetic makeup between individuals as contrasted to the effects of different environmental forces. For a review of the origin of the concept of heritability, see " Heritability in retro­ spect " by Bell (1977). The term heritability is used in both a broad and narrow sense. For the broad sense, the genotype is considered as a unit in relation to the environment. Genes come together in new combinations exhibiting intraallelic interaction (dominance) and interallelic interaction (epistasis). Heritability in the narrow sense considers only the additive portion of the genetic variability in relation to the phenotypic variability.
    [Show full text]
  • Title: Assessing Test-Retest Reliability of Psychological Measures: Persistent
    Title: Assessing test-retest reliability of psychological measures: persistent methodological problems Running header: Reliability of psychological measures Victoria K. Aldridge1, Terence M. Dovey2, & Angie Wade1 1Clinical Epidemiology, Nutrition and Biostatistics Section UCL Great Ormond Street Institute of Child Health, 30 Guilford Street London WC1N 1EH, UK 2Department of Psychology Marie Jahoda Building Brunel University Kingston Lane Uxbridge Middlesex UB8 3PH, UK *Author for Correspondence ([email protected]) Health and Life Sciences De Montfort University Leicester LE1 9BH, UK +44 116 2078158 This version of the article may not completely replicate the final version published in European Psychologist. It is not the version of record and is therefore not suitable for citation. Final published version of this manuscript: https://doi.org/10.1027/1016-9040/a000298 1 Abstract Psychological research and clinical practice relies heavily on psychometric testing for measuring psychological constructs that represent symptoms of psychopathology, individual difference characteristics, or cognitive profiles. Test-retest reliability assessment is crucial in the development of psychometric tools, helping to ensure that measurement variation is due to replicable differences between people regardless of time, target behaviour, or user profile. While psychological studies testing the reliability of measurement tools are pervasive in the literature, many still discuss and assess this form of reliability inappropriately with regard to the specified
    [Show full text]
  • Inference from Complex Samples
    Inference from Complex Samples By LESLIE KISH and MARTIN RICHARD FRANKEL The University ofMichigan and The University ofChicago [Read before the ROYAL STATISTICAL SOCIETY at a meeting organized by the RESEARCH SECTION on Wednesday, October 17th, 1973, Professor J. GANI in the Chair] SUMMARY The design of complex samples induces correlations between element values. In stratification negative correlation reduces the variance; but that gain is less for subclass means, and even less for their differences and for complex statistics. Clustering induces larger and positive correlations between element values. The resulting increase in variance is measured by the ratio deff, and is often severe. This is reduced but persists for subclass means, their differences, and for analytical statistics. Three methods for computing variances are compared in a large empirical study. The results are encouraging and useful. Keywords: CLUSTERS; COMPLEX SAMPLE; SAMPLING ERROR; DESIGN EFFECT; BRR; JACKKNIFE; INFERENCE; STANDARD ERROR; REPLICATION; INTRACLASS CORRELATION; SAMPLE DESIGN; SAMPLING VARIANCE; SUBCLASS ANALYSIS; STRATIFICATION; REPLICATION 1. INTRODUCTION STANDARD statistical methods have been developed on the assumption of simple random sampling. The assumption of the independent selection of elements (hence independence of observations) greatly facilitates obtaining theoretical results of interest. It is essential for most measures of reliability used in probability statements, such as aNn, chi-squared contingency tests, analysis of variance, the nonparametric literature and standarderrors for regression coefficients. Assumptions ofindependence yield the mathematical simplicity that becomes more desirable-and at present necessary-as we move from simple statistics such as means, to the complex statistics typified by regression analysis. Independence is often assumed automatically and needlessly, even when its relaxation would permit broader conclusions.
    [Show full text]
  • We Need to Talk About Reliability: Making Better Use of Test-Retest Studies for Study Design and Interpretation
    We need to talk about reliability: making better use of test-retest studies for study design and interpretation Granville J. Matheson Department of Clinical Neuroscience, Center for Psychiatry Research, Karolinska Institutet and Stockholm County Council, Stockholm, Sweden ABSTRACT Neuroimaging, in addition to many other fields of clinical research, is both time- consuming and expensive, and recruitable patients can be scarce. These constraints limit the possibility of large-sample experimental designs, and often lead to statistically underpowered studies. This problem is exacerbated by the use of outcome measures whose accuracy is sometimes insufficient to answer the scientific questions posed. Reliability is usually assessed in validation studies using healthy participants, however these results are often not easily applicable to clinical studies examining different populations. I present a new method and tools for using summary statistics from previously published test-retest studies to approximate the reliability of outcomes in new samples. In this way, the feasibility of a new study can be assessed during planning stages, and before collecting any new data. An R package called relfeas also accompanies this article for performing these calculations. In summary, these methods and tools will allow researchers to avoid performing costly studies which are, by virtue of their design, unlikely to yield informative conclusions. Subjects Neuroscience, Psychiatry and Psychology, Radiology and Medical Imaging, Statistics Keywords Reliability, Positron Emission Tomography, Neuroimaging, Study design, R package, Power analysis, Intraclass correlation coefficient Submitted 27 September 2018 Accepted 7 April 2019 Published 24 May 2019 INTRODUCTION Corresponding author Granville J. Matheson, In the assessment of individual differences, reliability is typically assessed using test- [email protected], retest reliability, inter-rater reliability or internal consistency.
    [Show full text]
  • The Intraclass Covariance Matrix
    Behavior Genetics, Vol. 35, No. 5, September 2005 (Ó 2005) DOI: 10.1007/s10519-005-5877-1 The Intraclass Covariance Matrix Gregory Carey1,2 Received 03 March 2005—Final 10 May 2005 Introduced by C.R. Rao in 1945, the intraclass covariance matrix has seen little use in behavioral genetic research, despite the fact that it was developed to deal with family data. Here, I reintroduce this matrix, and outline its estimation and basic properties for data sets on pairs of relatives. The intraclass covariance matrix is appropriate whenever the research design or mathematical model treats the ordering of the members of a pair as random. Because the matrix has only one estimate of a population variance and covariance, both the observed matrix and the residual matrix from a fitted model are easy to inspect visually; there is no need to mentally average homologous statistics. Fitting a model to the intraclass matrix also gives the same log likelihood, likelihood-ratio (LR) v2, and parameter estimates as fitting that model to the raw data. A major advantage of the intraclass matrix is that only two factors influence the LR v2—the sampling error in estimating population parameters and the discrepancy between the model and the observed statistics. The more frequently used interclass covariance matrix adds a third factor to the v2—sampling error of homologous statistics. Because of this, the degrees of freedom for fitting models to an intraclass matrix differ from fitting that model to an interclass matrix. Future research is needed to establish differences in power—if any—between the interclass and the intraclass matrix.
    [Show full text]
  • Psychometrics Using Stata
    Psychometrics With Stata Introduction Psychometrics Using Stata Chuck Huber Senior Statistician StataCorp LP San Diego, CA C. Huber (StataCorp) July 27, 2012 1 / 127 Psychometrics With Stata Introduction Humor involves two risks May not be funny May offend Disclaimer The use of clinical depression and its diagnosis in this talk is not meant to make light of this very serious condition. The examples use a depression index and the examples are silly (and fictitious!). This is not meant to imply that depression is silly. C. Huber (StataCorp) July 27, 2012 2 / 127 Psychometrics With Stata Introduction Given the current state of the world economy, the staff at StataCorp began to worry about the emotional well-being of our users. In early 2010, the marketing department hired a consultant to design a depression index to assess our users. The index was pilot-tested on StataCorp employees to determine its psychometric properties. The index was then sent to 1000 randomly selected Stata users. (All data are simulated and this is completely fictitious!) C. Huber (StataCorp) July 27, 2012 3 / 127 Psychometrics With Stata The Pilot Study C. Huber (StataCorp) July 27, 2012 4 / 127 Psychometrics With Stata The Pilot Study C. Huber (StataCorp) July 27, 2012 5 / 127 Psychometrics With Stata The Pilot Study C. Huber (StataCorp) July 27, 2012 6 / 127 Psychometrics With Stata The Pilot Study Administered to all StataCorp employees Explore the psychometric properties of the index Descriptive Statistics Item Response Characteristics Reliability Validity Dimensionality and Exploratory Factor Analysis C. Huber (StataCorp) July 27, 2012 7 / 127 Psychometrics With Stata The Pilot Study Descriptive Statistics .
    [Show full text]
  • Confidence Interval Estimation for Intraclass Correlation Coefficient
    Journal of Modern Applied Statistical Methods Volume 8 | Issue 2 Article 16 11-1-2009 Confidence Interval Estimation for Intraclass Correlation Coefficient Under Unequal Family Sizes Madhusudan Bhandary Columbus State University, [email protected] Koji Fujiwara North Dakota State University, [email protected] Follow this and additional works at: http://digitalcommons.wayne.edu/jmasm Part of the Applied Statistics Commons, Social and Behavioral Sciences Commons, and the Statistical Theory Commons Recommended Citation Bhandary, Madhusudan and Fujiwara, Koji (2009) "Confidence Interval Estimation for Intraclass Correlation Coefficient Under Unequal Family Sizes," Journal of Modern Applied Statistical Methods: Vol. 8 : Iss. 2 , Article 16. DOI: 10.22237/jmasm/1257034500 Available at: http://digitalcommons.wayne.edu/jmasm/vol8/iss2/16 This Regular Article is brought to you for free and open access by the Open Access Journals at DigitalCommons@WayneState. It has been accepted for inclusion in Journal of Modern Applied Statistical Methods by an authorized editor of DigitalCommons@WayneState. Journal of Modern Applied Statistical Methods Copyright © 2009 JMASM, Inc. November 2009, Vol. 8, No. 2, 520-525 1538 – 9472/09/$95.00 Confidence Interval Estimation for Intraclass Correlation Coefficient Under Unequal Family Sizes Madhusudan Bhandary Koji Fujiwara Columbus State University North Dakota State University Confidence intervals (based on the χ 2 -distribution and (Z) standard normal distribution) for the intraclass correlation coefficient under unequal family sizes based on a single multinormal sample have been proposed. It has been found that the confidence interval based on the χ 2 -distribution consistently and reliably produces better results in terms of shorter average interval length than the confidence interval based on the standard normal distribution: especially for larger sample sizes for various intraclass correlation coefficient values.
    [Show full text]
  • Statistical Methods for Research Workers BIOLOGICAL MONOGRAPHS and MANUALS
    BIOLOGICAL MONOGRAPHS AND MANUALS General Editors: F. A. E. CREW, Edinburgh D. WARD CUTLER, Rothamsted No. V Statistical Methods for Research Workers BIOLOGICAL MONOGRAPHS AND MANUALS THE PIGMENTARY EFFECTOR SYSTEM By L. T. Hogben, London School of Economics. THE ERYTHROCYTE AND THE ACTION OF SIMPLE HlEl\IOLYSINS By E. Ponder, New York University. ANIMAL GENETICS: AN INTRODUCTION TO THE SCIENCE OF ANIMAL BREEDING By F. A. E. Crew, Edinburgh University. REPRODUCTION IN THE RABBIT By J. Hammon~, School of Agriculture, Cambridge. STATISTICAL METHODS FOR RESI<:ARCH WORKERS-· By R. A. Fisher, University of London. .., THE COMPARATIVE ANATOMY, HISTOLOGY, AND DE~ELOPMENT OF THIe PITUITARY BODY By G. R. de Beer, Oxford University. THE COMPOSITION AND DISTRIBUTION OF THE PROTOZOAN FAUNA OF THE SOIL By H. Sandon, Egyptian University, Cairo. THE SPECIES PROBLEM By G. C. Robson, British Museum (Natural History). • THE MIGRATION OF BUTTERFLIES By C. B. Williams, Rothamsted Experimental Station. GROWTH AND THE DEVELOP:\IENT OF MUTTON QUALITIES IN THE SHEEP By John Hammond, Cambridge. MATERNAL BEHAVIOUR IN THE RAT By Bertold P. Wiesner and Norah M. Sheard. And other volumes are in course 0/ puJilication. Statistical Methods for Research Workers BY R. A. FISHER, Sc.D., F.R.S. Formerly Fellow oj Conville and Caius Colle/;e, Cambridge Honorary Member, American Statistical Assoc~f!tl0!1-.._~ .. Calton Professor, ziver~¥YA'l0 LOI!,ffn I 1,.'· I, j FIFTH EDITION-REVISED AND ENLARGED - ANGRAU Central Library Hydl'rabad G3681 /11.1111.1111111111111111111111 1111 OL~C~r;-.~OYD EDINBURGH:" TWEEDDALE COURT LONDON: 33 PATERNOSTER ROW, E.C. I934 }IADE; IN OLIVER AND :R~AT BRITAIN BY OYD LTD ., EDINBURGH EDITORS' PREFACE THE increasing specialisation in biological inquiry has made it impossible for anyone author to deal adequately with current advances in knowledge.
    [Show full text]
  • Estimation and Inference of the Three-Level Intraclass Correlation Coefficient
    University of Pennsylvania ScholarlyCommons Publicly Accessible Penn Dissertations 2014 Estimation and Inference of the Three-Level Intraclass Correlation Coefficient Matthew David Davis University of Pennsylvania, [email protected] Follow this and additional works at: https://repository.upenn.edu/edissertations Part of the Biostatistics Commons Recommended Citation Davis, Matthew David, "Estimation and Inference of the Three-Level Intraclass Correlation Coefficient" (2014). Publicly Accessible Penn Dissertations. 1252. https://repository.upenn.edu/edissertations/1252 This paper is posted at ScholarlyCommons. https://repository.upenn.edu/edissertations/1252 For more information, please contact [email protected]. Estimation and Inference of the Three-Level Intraclass Correlation Coefficient Abstract Since the early 1900's, the intraclass correlation coefficient (ICC) has been used to quantify the level of agreement among different assessments on the same object. By comparing the level of variability that exists within subjects to the overall error, a measure of the agreement among the different assessments can be calculated. Historically, this has been performed using subject as the only random effect. However, there are many cases where other nested effects, such as site, should be controlled for when calculating the ICC to determine the chance corrected agreement adjusted for other nested factors. We will present a unified framework to estimate both the two-level and three-level ICC for both binomial and multinomial outcomes. In addition, the corresponding standard errors and confidence intervals for both ICC measurements will be displayed. Finally, an example of the effect that controlling for site can have on ICC measures will be presented for subjects nested within genotyping plates comparing genetically determined race to patient reported race.
    [Show full text]
  • Guidelines for Reporting Reliability and Agreement Studies (GRRAS) Were Proposed Jan Kottnera,*, Laurent AudigEb, Stig Brorsonc, Allan Donnerd, Byron J
    Journal of Clinical Epidemiology 64 (2011) 96e106 Guidelines for Reporting Reliability and Agreement Studies (GRRAS) were proposed Jan Kottnera,*, Laurent Audigeb, Stig Brorsonc, Allan Donnerd, Byron J. Gajewskie, Asbjørn Hro´bjartssonf, Chris Robertsg, Mohamed Shoukrih, David L. Streineri aDepartment of Nursing Science, Centre for Humanities and Health Sciences, Charite-Universita ¨tsmedizin Berlin, Berlin, Germany bAO Clinical Investigation and Documentation, Du¨bendorf, Switzerland cDepartment of Orthopaedic Surgery, Herlev University Hospital, Herlev, Denmark dDepartment of Epidemiology and Biostatistics, Schulich School of Medicine and Dentistry, The University of Western Ontario, London, Ontario, Canada eDepartment of Biostatistics, University of Kansas Schools of Medicine & Nursing, Kansas City, KS, USA fThe Nordic Cochrane Centre, Rigshospitalet, Copenhagen, Denmark gSchool of Community Based Medicine, The University of Manchester, Manchester, UK hDepartment of Biostatistics, Epidemiology and Scientific Computing, King Faisal Specialist Hospital and Research Center, The University of Western Ontario, London, Ontario, Canada iDepartment of Psychiatry, University of Toronto, Toronto, Ontario, Canada Accepted 2 March 2010 Abstract Objective: Results of reliability and agreement studies are intended to provide information about the amount of error inherent in any diagnosis, score, or measurement. The level of reliability and agreement among users of scales, instruments, or classifications is widely unknown. Therefore, there is a need for rigorously conducted interrater and intrarater reliability and agreement studies. Information about sample selection, study design, and statistical analysis is often incomplete. Because of inadequate reporting, interpretation and synthesis of study results are often difficult. Widely accepted criteria, standards, or guidelines for reporting reliability and agreement in the health care and medical field are lacking. The objective was to develop guidelines for reporting reliability and agreement studies.
    [Show full text]