Over the years, market research has found it difficult to harness the power of the database through business intelligence tools like Cognos, Business Objects and CrystalReports. On the one hand, they seem like technical overkill: complex and time consuming to set up; on the other they dont go far enough, with their scant coverage for multiple response data, weighting and handling varying codeframes. The very data cube concept itself rapidly becomes a limiting factor, forcing arbitrary choices about what may or may not be cross-analysed. Omit something vital and a complete and costly rebuild is required, which often means a lengthy wait for the technicians to get to work.
Synola, a new MR-focused database query and report package, turns these assumptions on their head. Synola is a division of the UK-based Information Management Group, and its developers come from the slice-and-dice world of corporate database analytics and data cubes. Though not strictly an OLAP system, Synola uses OLAP tricks and techniques to allow researchers to approach their research data in the same way that the business intelligence gurus approach theirs.
The tools ability to aggregate data on screen may resemble the likes of Pulsar, StatXP, mrTables, Espri, or even, from an earlier generation, Quanvert at a superficial level. The difference is that it has no data file or proprietary data format behind it: the survey data always reside in an industry standard database, which could be anything from Oracle, DB2 or SQL Server to Access. Yet it does most of the things researchers need to do: query, tabulate, filter, re-classify, view charts, tables, apply significance tests and so on.
Speed is always an issue with database reporting tools: it is the problem that the pre-aggregated data cubes of most OLAP tools seek to overcome. Synola solves this with a blazingly fast accumulation engine that relies on some smart technology for handling bitmap data - our multiple response data - in the background. Its manufacturers claim a logarithmic performance curve, with the ability to process a million records in two-tenths of a second, ten million in two seconds and so on. By doing so, it avoids the need to build any cubes in advance: all queries are executed on the fly.
As a user, you pick the questions, or axes as they are termed from a typical cascading tree structure, to create a grid of data you are interested in, click the lightning-fork icon and your query is run. Now you can slice and dice, play with the data, post it to charts and so on, or go back and build another query, There will be a terminology jolt for most MR users, and some of the terminology (member for a pre-coded answer to a question) needs improving.
When selecting for any query, the top level of the tree represents all the different surveys available, which marks another significant breakthrough: a solution that offers one database with all your surveys on it. Queries can span surveys, which is particularly relevant to panels or to continuous customer satisfaction work where respondents are re-interviewed: all their data can be unified in this.
|Questions too can be multi-dimensional. A battery of attitude scales thus become a single object, and on a tracker, time could add yet another dimension. In your query, you assemble any number of one, two or multi-dimensional objects, then click on a lightning fork icon to execute the query and present the results.
There is an impressive array of presentation options, including some very clear, multi-dimensional charts. There is a wonderfully intuitive and visually appealing filter definition tool that uses coloured venn diagrams to illustrate and apply your selections without needing to think about ands and ors.
There are some major lacks at present, though. Weighting is only supported with respect to mid-point scale values which must be built into the database and cannot be user specified. There is no respondent weighting, and the handling of means and related descriptive stats is too limited. Labelling on tables is also rudimentary and needs to improve for reporting purposes. However, in my discussions with the developers, they demonstrated a willingness to respond quickly, and actually implemented Chi-square significance highlighting in the couple of weeks between my first viewing of the product and my second. It seems likely these other gaps will get plugged fairly soon.
Adding surveys to a relational database can be onerous, but with support for SPSS MRs data model and triple-s coming, this will get easier. The biggest challenge will be to shift MR to the relational database paradigm. Databases tend to sit with IT, not DP, so for many, the present mish-mash of files avoids ever having to face up to these strategic issues. But, for the consumers of research, the continued failure to get research data into the corporate knowledge network is leaving MR vulnerable and marginalised. This could be the crossover product the industry needs.