Simulator Setup Files

Metrisim can automatically create an entire conjoint analysis simulator from a few csv files that you provide at the start of a project. When logging in you will be asked for at least the essential files and some optional files. Tip: take a look at the demo files to better understand layout, and also read through this article get a better understanding of how the system works.

Essential files include:

attribute_map.csv your map of attributes and levels.
pws.csv partworths and weights. NOTE: This can now be automatically generated by Metrisim if you use the full suite.
control.csv contains your baseline scenario (the current market or your best approximation of it).

Optional files include:

The targets.csv contains the actual market shares that can be used by the calibration function in the simulator to weight shares so they more closely align with actual shares. If volume weights are used, then this is share of volume, if not, then it is interpreted as share of choices. In a durables or subscription market, share of choice is usually adequate as long as you expect each consumer to usually buy one unit in the survey’s defined time horizon (e.g. espresso machines in the next 12 months). The file is optional, as it is: 1.) usually preferable to instead explicitly model external effects using a set.csv file; 2.) market share data might not be available. However, either a targets.csv or set.csv file should be used to more closely approximate actual shares, unless the goal is to exclude external effects, such as promotion and accessibility.
The set.csv contains binary 1 or 0 indicators of product presence or absence in the respondent’s relevant set. The relevant set includes brands that the respondent is aware of and that are available to the respondent to buy.

1. Attribute map file

Your attribute_map.csv file must contain the product attributes and levels they are mapped to. A column heading is an attribute and each row below is a level.

DO NOT reuse level names within an attribute or across attributes, each must be unique.

Price attributes should be labelled as ‘Price’ to be detected so that partworths can be interpolated. Prices should not include currency symbols.

Attribute names and levels should be indentical across all the setup files (attribute_map.csv, control.csv and pws.csv). Preferably stick to letters, numbers and spaces. If you must, the only special characters allowed in the simulator include: commas (,), ampersands (&), colons (:), bracket (()) and percentage symbols (%).

Levels, with the exception of 'Price' must not start with a number. Attributes must not start with a number. The word None is reserved and cannot be used as an attribite level; alternatively use words such as 'Nothing' or 'None offered'.

Figure : Screenshot of a Metrisim attribute map file (attribute_map.csv)

2. Partworths file

If you don’t have software to estimate utility partworths, the open source R conjoint package is recommended.

The pws.csv file contains partworths (mean-centered regression coefficients estimated per respondent) for each attribute level (one column per level). See the provided demo file. Keep the order the same as in the map file. Be sure to use the standard name for the file, pws.csv.

Please ensure the probability ratings used to estimate these partworths range between 0 and 10. For instance the Juster scale or Verbal Probability Scale conform to this. If not, you can rescale them to conform. For instance you may have 0 to 100 ratings, rescale them to conform by dividing by 10.

Figure : Screenshot of the first few rows of a Metrisim partworths (pws.csv) file

2.1 Price partworths

The price partworths should have numerical column headers –no currency symbols are permitted. Even though you will only have tested a few price points, Metrisim uses interpolation to take care of other possibilities. Interpolation allows you to obtain partworth estimates for prices ‘between’ the tested price levels, during simulations. For instance, if you test, 5, 10 and 15 but you want to run a scenario with a price of 12, the partworth will be interpolated for each respondent. However, ’extrapolation’ is not possible, as no information is available regarding the shape of the curve beyond the tested price range.

2.2 Weights

You will notice in the demo pws.csv a Weights column. This is by default, 1 per respondent, which is effectively the same as having no weighting. You can instead insert either: respondent weights (to improve representation if needed - usually calculated by a RIM weighting package such as Anesrake) or volume weights. You may also have a compound of the two. Volume weights might be needed in a market where consumers usually purchase more than one unit over the forecast period. The volume weight will affect the share % calculations - especially if volume correlates with partworths. For example, if the buyers who choose your product in a competitive simulation also tend to be the type of buyer who purchases more units than average, then your product's share should be higher.

Keep in mind that you will need to adjust RPK (revenue per 1,000) provided by the simulator in a CPG or other market where you expect more than 1 unit to be bought in a period. The simulator estimate is based on the simple formula (share % / 100) x 1,000 x price, which in effect is assuming 1 unit is purchased by each consumer over the forecast period. So for instance if the average is 110 units, you would then multiply the RPK by 110 to correct for this and display that in your presentation instead. If there is sufficient interest, Metrisim may include a more automated way of doing this in future. If you like this idea, please reach out via the suggestions button at the bottom of the webpage.

3. Control file

The control.csv file defines your baseline scenario (current market scenario) for reference. When you press ‘reset’ on the control panel it resets to this. Try to approximate the current market as best you can with the available attribute levels. The first column is headed ‘attributes’ and must contain one attribute per row. The product names (column headers) to the right could be named sequentially, such as p1 onwards (for each product expected in your scenarios). For instance: p1, p2 and p3. Alternatively, you could use unique text labels to aid recall. Remember though that in conjoint analysis products are defined entirely by attribute levels selected in the simulator, not by the product labels.

The second row is an inclusion indicator. Usually set all to Yes. In simulations users can then decide whether to include or exclude specific products by selecting Yes or No from the dropdowns. Subsequent rows contain levels. Make sure these exactly match the wording using in your attribute_map.csv.

Figure : Screenshot of a Metrisim control.csv file containing the baseline scenario

4. Targets file

Optional file

The targets.csv file is used to provide either actual market shares or calibration weights to the system. If you provide actual market shares, Metrisim automatically calculates calibration factors (the simulator makes use of the control.csv to provide the baseline scenario - which is assumed to match the actual market). The automatic calibration against actual market shares only makes sense if shares are available for each product - it won't make sense if you are introducing a new brand / product without a sales history.

If you provide calibration weights manually, then ensure they are a multiplier indicating the amount by which the predicted share should be multiplied. You can also use this if you only want to calibrate the none option - by setting all competitors to 1 and only supplying a multiple for the none option. Ideally though, it's best to only use calibration as a last resort. It is better to model external effects explicitly using the set.csv file.

See below for more detail:

4.1 All products are already on the market

Conventionally actual market shares for the baseline scenario (current market) would be included in the targets.csv. You could estimate market shares using purchase behaviour questions in your survey questionnaire or use an external source. Just ensure the type of market share aligns with the output of the model (see the section on respondent weights).

The calibration weights are estimated automatically against the baseline scenario and the shares are multiplied by these weights. They are normalized to 100.

Figure : Screenshot of a Metrisim target.csv file (of type "target share %s")

4.2 New concept vs existing products

Sometimes the aim of the conjoint analysis is to determine the features and pricing of a yet to be launched product. In this case, no market share data will exist. So what can you do? One option is to instead use the relevant set method provided by Metrisim (see the set file section).
However, if you prefer to use calibration, then you can provide manually calculated calibration weights in the targets.csv file. Metrisim will automatically detect this - by assuming that if the average of the values is below 2 - you are providing weights instead of shares. You will however have to determine a way to estimate a reasonable calibration weight for the new concept to reflect the amount of marketing support you expect the new product to receive at the time of launch up to some reasonable forecast horizon. For instance 1 year ahead is common in FMCG.

Figure : Screenshot of a Metrisim target.csv file (of type "manually calculated calibration weights" instead of target share %s)

4.3 None calibration

Usually the none option should be calibrated, no matter the type of conjoint analysis, CBC / DCM or other, there is a tendency for the none percentage estimated by the model to be too low. One possible cause is that the researcher designed the conjoint analysis exercise to only cover a limited subset of brands, as though they the consumer were in one store. However in the real world, consumers know they can go to another store if they have a strong preference for a brand that is not in stock. In cases where they don’t have a strong preference they may easily switch brands, and so if the design excludes a common brand, they may simply buy another, while in the real world they would usually find it on shelf.

To calibrate the none option, provide manual calibration weights in targets.csv. Manual calibration weights are a multiplier indicating the amount by which the predicted share should be multiplied. Set all products to 1 - meaning you are not calibrating the products - and supply a multiple for the none option.

5. Set file

The set.csv file contains indicators – either a 1 indicating a product is part of a respondent’s ‘relevant set’, or 0 indicating it is not part of the set. For instance a product that is not available where the respondent shops is not part of their relevant set. There are other factors which may also play a role in preventing purchase, even when the respondent has a favourable opinion of the product.

Figure : Screenshot of a Metrisim relevant set file (set.csv)