Digital Workshop

I have been asked to produce a program that requires users to answer a series of questions relating to the likely outcome of their business relationships.

For example, they will be asked how long they have known a client, whether or not they enjoy a close relationship, whether the product/service they are aiming to sell is relevant to the client’s business, whether there is a stronger competitive product etc.

The aim is to supply a forecast at the end of the process – something like “Based on your current relationship with the client in question, and taking into consideration the other factors you have identified, there is a 72 per cent likelihood of achieving a sale.”

Each question/answer will be given a percentage value and, obviously, (adapting the scoring function in Opus?) scripting will be required to produce the final percentage figure.

At which point I’m beginning to run out of steam. Can anyone suggest the best way of delivering what’s required?

Regards

Roger

Hi Roger, I'm wondering about this - can you explain how the weighting might work - the maths?

How many options/responses would the user have per question? The same number of responses per question (A:always, E:never sort of thing), or might there be different numbers of responses - eg some yes/no's etc...

For example:

Here's a question: Do you speak to the business by phone?

The answers may be: Often, Sometimes, Never.

If the user picked each of these three responses, how might it affect their final percentage? Mathematically-speaking.

Is there a formula? Presumably these questions and weightings are based on some stats that have already been collected. They're probably in tables or an excel worksheet - if so, I'm thinking a database with a set of weighting tables and some scripting would do it.

Don't know if this is of any help...



Melanie

Hi Melanie

There may be different numbers of answers to each question (three, though, would be the average). The intention is that each answer would be given a percentage weighting - e.g. 80% for the first answer, 60% for the second, 40% for the third...

I believe there is an Excel formula relating to this, but my lack of Excel knowledge propels me away from making use of it. In view of the fact that the maths would seem to be fairly simple - the percentage score for each question being divided by the number of questions at the end of the process to give an overall percentage score - I'm leaning in the direction using the Opus scoring function.

The problem is (a) that I've not used scoring before in any of my Opus publications and (b) that my scripting ability is limited. Any chance you can point me in the direction of the best way to proceed?

I appreciate your response.

Regards

Roger

Hi Roger
I just wanted to say, that your method of maths would be fine if each question is weighted equally, however when you are trying to ask a series of questions in order to predict an outcome or liklihood then you have to carefully consider predictive validity.
You may well find that some questions are far better predictors of an outcome or liklihood than others and if this is the case you'll want to weight those higher than other questions.
For example, if you were trying to predict the liklihood that someone will have a car crash in the next twelve months, I would think a question such as "Do you drink and drive?" will have much more predictive validity than a question such as "Do you play your car radio while driving?"
Unfortunately, I have forgotten most of the stats subjects I did many years ago at Uni, but if you can have a chat with someone who is up to speed on stats you will probably increase the chances of your questionnaire actually predicting what you are trying to predict.
Cheers
Graham

Hi Roger

I've set up a few quizzes and questionairres in Opus.

In test construction, questions (or statements) are preferably researched on a representative population, producing correlations that then allow for some predictive validity. If the formula available to you has been derived in this manner, then it's likely all answers will, as discussed in other replies, not be equal and will need to be weighted to account for this variance.

On the other hand, if the formula is not based on research but informally created, (for example, a score of 80% predicts a "high" likelihood of an event's future occurance), then your task may be alot easier. If this is the case, you can use a Likert scale which has 5 answers, ranging from always, frequently, sometimes, rarely, never or any of a number of variations on this theme. Always correlates with a full weighted score. I haven't used percents, rather I give always a 4, frequently a 3, sometimes a 2, rarely a 1 and never a 0. These can be reversed for questions where a full weight is given to a negative answer, so that always is given a 0, etc. So, as an example, for "I always get prompt attention for serious health problems," an always would be scored a 4 and a never would be scored a 0. If worded "negatively," for "I never get prompt attention for a serious health problem," if that's "always" the case, that gets a 0.

In Opus, an easy way to score the answers is to set a score (set variable action) based on 0-4, for each answer, storing it in a separate "answer" variable for each question (for example, "answer1," "answer2," etc.). You can require that each question is answered by not allowing the page to forward to the next page and question until a response is selected (can script something like "if answer1==null show. text box//words like please answer all questions// else, GoToPage.pagenumber").

Then on a total score page, a script like answer1+answer2, etc. If the score range is more intuitively created than normed using research and a representative population with resulting formula: then predictions are hypothesized based on each range of scores. Not very scientific but, absent an empirically verified formula based on rigorous research protocols, what you have.

If this is not too muddy and remotely useful, and you'd like to inquire further, feel free to send a private message. Happy to try to help further.

Kind Regards,
Stephen

Thanks for the input, Graham and Stephen. I anticipated that the problem is likely to involve more than simple arithmetic and, armed with your assessments, I think I'll go back to my client and chat with him about "predictive validity".

I appreciate the offer, Stephen, and - depending on my client's reaction - may well take you up on it.

Kind regards

Roger