`Home     Our Services     Your Resources     News & blog     About us     Contact

Marsc 5.0 Reviewed

Tim Macer reviews a program that can create quota-based or convenience samples from panel systems or enterprise, sample, and CRM databases

Statisticians have their own reason for calling the randomly drawn, quota balanced samples of most MR surveys ‘convenience samples’. Yet convenient is not the word most likely to spring to mind as you attempt to pull a balanced internet sample from a panel database. The complexity of the selections can spiral out of control as selection criteria, often spelled out in arcane SQL queries, not only attempt to fill interlocking quota cells by demographic, as well as randomly and evenly, but stretch to take into account past response history and even propensity to respond. Draw too much sample, and too many of your hard-to-get respondents will be booked out for this one project and unavailable to other projects about to launch; fall short, and costly, time-wasting top-up samples loom.

When Interface last reviewed Marsc in June 2003, it was aimed at enterprises feeding samples from multiple CRM systems to agencies largely for just-in-time customer research. It will still fulfil this role, but now, with Marsc version 5 (and with 5.1 launching this month), the focus is on research and fieldwork agencies running their own panels.

The system consists of a database of contacts, a sample selection interface that you use to draw samples and produce reports on sample availability and usage, and several associated processes to extract data, and importantly, to update your sample with interview outcomes, as the survey progresses. This completing of the loop is vital, as it allows you to build up response history for each panel member. Marsc will then take this into account when selecting samples, as it is able to calculate a response probability for each demographic cell in your quota matrix based not on speculation but on actual past history.

It is the main sample selection screen which is the where the power and flexibility of the tool come together in a fairly intuitive way to any researcher or project director. It steps you through a series of stages in extracting a sample from the database. First you select who you do want, with an option to produce an interlocked model, where you have a target for each cell in your quota matrix, or a non-interlocked model, where you only specify the targets for the totals in the margins of your quota matrix, and the system then applies a regression-based statistical model to estimate the best fit across the whole table, and draw sample accordingly.

At the next level, you define exclusions or filters, which could be participation in other related surveys, or time-related resting rules on recency of participation, as you could before. New to this version is the ability to select positively on respondents who have not recently participated, in order to work the whole panel, and not just an active minority.

The tool’s sampling credential are impeccable from a methodological point of view - even to the extent that the ‘sample’ is considered to be those who actually participate, not those selected for interview. Any targets given for sample size always relate to the number of completes required. Response rates, either historical or estimated, are then applied as a projection factor to select the contacts.

The new Marsc will either work as a complete panel management system, or alongside existing panel solutions. As a standalone system, its own database is capable of expanding both vertically to accommodate panels of limitless size, and horizontally, to include as much profile or demographic data as you wish, or to add to this incrementally.

There are some panel administration capabilities built in, including tools to build the panel member interface for them to manage their own profile and check and redeem their rewards. However, this new module is still fairly raw and does not have the depth of functionality of the core selection, and would probably need a fair amount of customisation if used for a real panel. Another curious omission is the invitations and reminders capability needed for internet surveys. Perhaps Centurion assume that this would be done by whatever web survey tools is used, but there are often good operational and even data protection reasons for keeping invitations at arms length from the survey component.

With respondents becoming still harder to get, and panels continuing to rise in popularity, this is a tool that will be hard for research and fieldwork companies to ignore, in the running battle to sample smarter. 

The user view: Kevin Goldbert at Research Now

Kevin Goldberg, Director of Development and Panels at Research Now, a UK-based European internet Fieldwork Company, is currently introducing Marsc. It will integrate tightly with the company’s existing and extensive panel databases. For him, the prize is both more effective use of the panel, and giving project managers more direct control.

“Previously we were running SQL queries on our panel database to select sample,” reports Goldberg. “These were highly complex and they could potentially take a long time to write and run. Even though it was a partially automated process, it was not especially efficient in terms of the set-up time. I liken the approach in Marsc to rim weighting: you put what you need at the edge and let the cells work themselves out. You have better re-use and also more precision. It lets you structure the sample precisely to the survey quotas and it also uses previous history to determine the response: though you can over-ride that if you require. This means more efficient use of the panel and less wastage, but also means you are not likely to run out of sample either.”

Goldberg considers this ease of use is vital in shifting the focus of the task from an expert SQL programmer to users who have research skills. He says: “A researcher that will then be thinking about response rates and feasibility – as it is quick and easy to see who is available. It also gives more visibility to where potential shortages may be occurring, so by deskilling on the SQL side, you can take a more intelligent view of your sampling operations that goes beyond the immediate project. It means that more people will be able to make panel selections. Now we have new project managers in the US and in Australia and the good thing is that they will be able to use it by themselves at times to check feasibility or draw samples when our UK-based sampling team is in bed.

“Early on, when you have a few projects, something like this does not seem so important; but to handle the volume of projects and number of panels that we do the level of complexity becomes huge, and you then realise you need to have something like this.”

Published in Research, the magazine of the Market Research Society, April 2006, Issue 479

© Copyright Meaning Ltd 2006. All rights reserved. Reproduced with permission.

Assists with the random selection and creation of quota-based or convenience samples from sample databases, panel systems or enterprise and CRM databases with a built-in panel database and panel management capability, if required..

Ease of use
Cross-platform compatibility
Value for money

From £8,000 for single user, sampling only; £15-£20,000 for a multi-user panel solution.