This paper describes the use of computerised topic analysis to segment and analyse tweets by US presidential election candidates in 2016. The statistical technique used to create the topic is called Latent Dirichlet Analysis (LDA). The paper shows how LDA can be used to automatically generate topics un textual data and discusses the potential to use LDA as both a static 'batch" and real time analysis approach for textual data.
The web provides researchers with a gold mine. Every click, response, page movement and delay can be logged and used for analysis. The amount of data that can be captured from web surveys is vast and comprehensive. This paper aims to explain the reasons why Paradata should be collected and analysed. Paradata is a vital partner to the other sorts of data from surveys, response data and metadata. It provides a valuable tool to both compensate for inadequacies of web surveys and provide an advantage that web surveys have over more traditional surveys.
The core reason for this paper is that Web surveys bear little resemblance in terms of their execution to the current dominant methods of survey data collection. In the United States, CATI (computer-assisted telephone interviewing) still predominates, although rising refusal rates are making it increasingly hard. In Europe, CAPI (computer-assisted personal interviewing), CATI and paper-based face-to-face interviewing are all widely used, the dominance of each method varying across countries and business sectors. In the Pacific Rim overall face-to-face paper-based interviewing predominates. CATI is also used where the telephone infrastructure, demographics and cultural acceptability permit. In CATI, CAPI and paper-based face-to-face interviewing (the acronym PAPI - paper-assisted personal interviewing - will be used) there is a human mediator or interface to the survey instrument. The interviewer presents questions and records the responses. Self-completion surveys, while used in many business sectors, arc not a dominant mode of data collection.
Web surveys are now an established part of market research. However one area that has not been closely studied is the behaviour of respondents in terms of their interaction with the questionnaire. Web surveys are self- completion surveys unlike traditional CATI and ('AIM surveys. Respondents complete the survey in isolation and the exact nature of the user interface varies considerably. Browser options can vary what the respondent sees radically and the survey designer has little control over this. Using logfiles from a commercial Web interviewing system (scy Web) actual respondent behaviour in the interview and errors can he tracked and recorded. This paper presents an analysis of (his data from several different surveys. Question types and formats that seem to cause the respondent either to quit or wrongly complete can be identified via these logfiles. The author attempts to develop a taxonomy of respondents based on their behaviour monitored via the logfiles. Finally the author puts forward some proposals for the enhancement of Web survey tools taking into account their unique nature.
The paper describes the implementation of a system using a CATI package and a predictive or statistical autodialer. The autodialer has the capacity to detect if a human voice has answered the telephone and make a connection to an available interviewer, this also means that busy signals, ring no answers and other unwanted calls can be eliminated. Dialing many numbers at once means that interviewers can be kept almost continuously interviewing. The principles of predictive autodialers are described with reference to their enhanced performance over manual dialing systems. The construction of the system interface between the EIS autodialer ( Electronic Information Systems Inc USA) and the Quancept CATI package is examined next. This describes the form of changes required to the CATI system (i.e no longer user initiated dialing) and how this can influence the performance of the package The next section of the paper describes a "case history" of such a system. This reviews management and performance issues which occurred in a real life situation using the autodialer system described previously. Finally there is a discussion of future work on the system that is planned and speculation on the impact of such systems on the economics of telephone interviewing.