Self-completion gets a character reference
Formic and Readsoft DOCUMENTS (formerly Eyes & Hands), January 1995
If the research you do involves self-completion questionnaires, then be prepared for a revolution in the way you work over the next few years. That is the message coming from the "early adopters" of the latest Optical Character Recognition or OCR-based questionnaire scanning systems. Not only have the questionnaires for scanning started to look more like "real" questionnaires, unlike the earlier optical mark recognition or OMR-based systems, but users are now claiming they are achieving a better response rate too.
OMR is a safe and reliable system, having been used for many years. For instance, all the TGI data for BMRB is scanned in using the DRS OMR system. OMR is the system being used for the new National Lottery. It works well in high-volume situations but makes for eccentric-looking forms that are not always so easy for respondents to complete.
Leslie Sopp is the Head of the Consumer Association's survey centre. They installed Formic last summer to automate the input of their subscriber surveys they do for the various Which? titles. "It has revolutionised the work we do" he told Research. "I would say this is the equivalent of the CATI and CAPI revolution, only this time for self completion questionnaires. Previously, we had three full-time staff and 12 temps, which was very expensive. We can now handle all the self completion work with just three staff". Their recent widely reported survey on banking was one of their first scanning projects. They received back 4000 of the 8000 12-page booklets they mailed out. Scanning and processing took them just three weeks - less than half the time of comparable jobs when punched.
Formic is a design and scan system, with its own word-processor-style module that you use to create a special scanning questionnaire on screen. The questionnaires it produces are presentable enough for self-completion work such as panel surveys, although probably not for high-profile end-user surveys. Questions are entered into frames along with tick-boxes in a straightforward point-and-click way that will be intuitive to any Windows user. For openended questions, a special frame is created which will be picked up when scanning then coded on-screen. Numeric responses can be dealt with either in this or as a number-grid, which, though automatic, are confusing to respondents.
It contains a number of neat features such as choice of type styles, automatic question numbering and easy alignment of boxes, but lacks others such as the ability to add logos or arrows, or any automatic support for routing (surely a necessity if questions get renumbered automatically?).
Reading the forms is pretty foolproof too. You just load in all the questionnaires that the morning's post has brought - even if they belong to different surveys, and they will be scanned and sorted into the right "database" for each survey. This is the point, according to Keith Negal, Director at Formic when "you can throw away the scanned questionnaires, since you now have an accurate scanned image of the form on disk, like an electronic photocopy". There is then good support for coding numerics and openends on screen.
Eyes & Hands from Readsoft, although not yet as established in this country as Formic, has much in common with it and is, from the operational point of view, very similar. A major difference is that it has a "set-up" module, not a design module. You actually scan in a blank questionnaire which you will have designed outside the system. This has the virtue of allowing you to read questionnaires that were created or printed externally. The set-up makes light work of marking up a blank questionnaire - you define the questions with a few mouse clicks by pointing to the scanned page on screen.
Eyes & Hands will read handwritten characters too. At present this is confined to numbers, although it is not necessary to write each digit in a separate box. Sandy Gellan, MD of Readsoft, says their ICR capability is based on a proprietary algorithm which has enabled them to develop a recognition system which is extremely accurate and tolerant. He says "we will have support for uppercase letters too, in our next release". This would make it possible to read and validate postcodes and determine acorn groups against a CD-ROM copy of Royal Mail's PAF (Postal Address File).
RSL uses both Eyes & Hand and Formic. They annually scan 144,000 questionnaires for the RAJAR survey using Formic. Each questionnaire is tailor-made to the respondent on the RAJAR panel, with key information pre-printed on the form. There are also regional variations with questions or sections omitted. Where the questionnaire is designed externally, or printed in colour, they use Eyes and Hands. Roger Calverly, Director at RSL, dismisses any concern over mis-reads being overlooked. "We have done a lot of tests since we started and we believe we get better quality data from OMR than we do from data entry."
Group Sigma in London, who are data entry specialists, chose Eyes and Hands to increase their throughput of work. Mike Safey, MD, also spoke in revolutionary terms. "In five years' time" he predicted, "people are going to be wondering how they managed without this. To us, it is the best thing since the proverbial sliced bread". The previous weekend, two operators had scanned 14,700 A3-sized questionnaires, which Safey estimates would have been three months' worth of punching. They achieve total accuracy with tick-boxes and numbers and are reaching 85% initial acceptance on written-in numbers - they key the other 15% from a snapshot on screen.
All the users I spoke to had initially experienced some problems with getting the readers to read what was on the questionnaires. Two factors seem to be at work. First is the design of the questionnaire which can affect both accuracy in completion and readability through the scanner. At the Consumers' Association, the person who will eventually handle the processing of the survey is now the one who has the final say in the layout of the questionnaire. Second is adjusting the tolerance and sensitivity of the reader. There were several reports of jobs having to be rescanned because the sensitivity was too low and half the data was simply ignored, or it was too high and the reader started reading empty boxes. Fortunately no-one actually had thrown the questionnaires away after reading them so it was simply a matter of rescanning. It did demonstrate that greater vigilance is required during processing before accepting the questionnaires as "read".
Judging by the endorsements these products have received from those using them, and the fact that much of this enthusiasm is due to the considerable savings they are making in both cost and time, it would be remarkable if we did not see these systems being snapped up in 1995. And since OCR services are now being offered on a bureau basis, you could still invest in the technology without having to invest in the hardware.
Published in Research, the magazine of the Market Research Society, January 1995, Issue number 344.
© Copyright Market Research Society/Tim Macer 1997. All rights reserved. Reproduced with permission.
Copyright © 2006-2009 meaning ltd. All rights reserved.