Phil Hearn: Blogger, Writer & Founder of MRDC Software Ltd.

# What is rim weighting – with free Excel working model

Rim weighting is a technique commonly used to weight market research data to known targets – e.g. age groups, regions and gender. The technique will allow you to independently weight data to each variable (question). MRDC can provide you with a free working model in Excel please click here to download.

### Scope

The article only explains how rim weighting works and some basic checks you should apply.

### What is rim weighting?

Rim weighting is a particular form of target weighting. It can be a practical tool when you have targets (or populations) to which you wish your data for two or more variables but not targets for the interlocking cells for these variables.

For example, you may know that your target sample should be:

Males – 50%

Females – 50%

16-34 – 40%

35+ – 60%

These targets are known as ‘rim weighting targets’. You may have more than two variables, where rim weighting is likely to be your chosen method. However, read the notes in section 4, as you should not see this as a panacea for all sampling errors.

### Target weighting with interlocking cells

If you do not know the percentages or actual figures for the interlocking cells, you will not be able to use ‘standard’ target weighing. To use ‘standard’ target weighting, you would need to know the following, for example:

Males 16-34 – 15%

Males 35+ – 35%

Females 16-34 – 25%

Females 35+ – 25%

Generally, the targets for interlocking cells are better than rim weighting. However, where several variables comprise the targets or there are many items within each variable, ‘standard’ target weighting may be inappropriate or impossible to apply.

### How does rim weighting work?

Rim weighting works by what is known as an iterative target weighting process. In other words, the software (assuming it can perform rim weighting) will calculate targets for the first rim. In the example above, the process would apply weighting factors that would achieve 50% males and 50% females.

After applying this weighting factor, it is highly improbable that the targets of 40% for 16-34-year-olds and 60% for 35+-year-olds would be achieved.

The software would, therefore, calculate a multiplicative weight that would adjust the data so that 40% for 16-34-year-olds and 60% for 35+-year-olds is achieved. Applying this multiplicative factor would almost certainly mean that the targets for males and females would no longer equate to 50% each.

### Rim weighting calculators converge on the targets

Now, the iterative process begins. The software would now apply another multiplicative factor so that the gender was weighted to 50% males and 50% females. Then, it would re-weight to 40% 16-34-year-olds and 60% 35+. And so on and so on. The multiplicative weights converge on the desired target, giving you weighting factors to apply to achieve the desired targets.

As the program performs the iterations, the data gets closer and closer to the desired targets. In some cases, it may be impossible to reach the exact targets. Most software programs that can handle rim weighting will have a fixed number of iterations it will attempt before it gives up. In some cases, due to the structure of your sample and the laws of mathematics, it may not be able to achieve your desired targets; in some cases, it may be utterly impossible as your sample is so far skewed or biased to be able to reach the targets you are seeking. There are some notes in Section 4 about this.

Where more than two variables are used as rim weighting targets, the iteration process will pass through each variable in turn before it starts again at the first one.

Please see this video if you want to know how to carry out rim weighting in MRDCL.

### Checks you should apply and reasons for caution

There are checks and cautions that you should consider (in no particular order of importance):

• ##### Always check the factors produced

Rim weighting is more susceptible to volatility than ‘standard’ target weighting, particularly when you have more than two rim weighting variables. It is essential that you check the range of weighting factors that are being applied to each record rather than being satisfied by the fact that the output looks right because it has met your desired targets. As a rule of thumb, the lowest factor should not be less than half the average factor applied or more than twice the average factor applied. However, you should consult a qualified statistician if you have concerns about the legitimacy of any factors used.

• ##### Avoid too many rim targets

Rim weighting will prove particularly volatile when you apply too many targets. If you are getting volatile weights or some outliers, this may be caused by having too many targets. The program may apply a very high weight to achieve one or a small number of your rim targets.

• ##### Take care with ‘skewed’ variables

Rim weighting has the propensity to produce volatile weight factors when your targets are highly skewed. For example, setting a rim target of TV Viewers (98%) and non-viewers (2%) may cause volatility. If you have several variables skewed in this way, the results are even more likely to be volatile.

• ##### Variables with too many responses need care

Rim targets with too many items may cause volatility. For example, if you make a list of 25 publications that a respondent reads most often as one of your rim targets, this is likely to stretch the data too far. Unless you have a large sample, grouping the publications into publication types may be better.

The smaller the sample you have, the fewer rims you should have. You should seek a statistical expert to advise on this. A sample of 500, for example, would usually be fine for 3 rims of, say, 2, 3 and 4 items.

• ##### Do rim targets have high correlations

Rim weighting may not work where variables are highly correlated. For example, if you set rim targets for those with a high/medium/low income and for owners of luxury/midrange/cheap cars, there would probably be a high correlation between luxury car owners and high-income respondents.

### Check your effective sample size

After applying any type of weighting, it is good practice to check your effective sample size. Our software products have this built into the software as an option. The effective sample size will always be lower than your actual sample size. The effective sample size tells you the equivalent number of respondents you could have interviewed with perfect sampling. If, for example, your effective sample size drops from 1000 respondents to 400 after rim weighting, this means you could have sampled 400 respondents to match your criteria and achieved the same reliability in your data.

### Need a working example?

Try our working example in Excel, which will calculate rim weighting factors from the targets you input. If you need software that can handle rim weighting and produce crosstabs, please contact nikki.sunga@mrdcsoftware.com. We also offer consultancy and advice if you need help with any of the techniques discussed in this article. You can download the calculator here.

### More to explore

##### How to analyse hierarchical data in market research

There are two types of hierarchical data encountered in market research. These are respondent-based hierarchies and data-based hierarchies. In practice, they are analysed similarly, but, more importantly, they need software capable of analysing hierarchically structured data.

##### Is Quantum Still The Best Crosstabulation Software For Market Research?

There are around 100 software solutions for collecting market research survey data online. Many of these 100 solutions have compelling reasons to choose them instead one of the other 99 offerings.

##### How to learn a survey scripting language like MRDCL

Market research software is in some ways no different from other software, but you are likely to need and understand detailed features for survey data collection, analysis and reporting. These detailed requirements mean that you will want to care about productivity, best practices and, often, learning the full scope of any software product