ChatterBank4 mins ago
Statistical Methods
Can someone please help an aged brain?!
Many, many moons ago when I was at university, one of my close friends was studying statistics and of course we talked about or respective subjects.
One of the things that he mentioned was a method allowing probability/statistics to help produce a set of variables to try and predict an outcome.
Probably not explaining this well but I now need that sort of algorithm at work but I cannot remember the name of the methodology.
Can someone help?
Basically I have a multi-variable input scheme (about 6 inputs) that vary in value from 0 to an upper limit (each independent and known).
I want to be able to analyse these variables over their input range to produce an output set where 90% of the results of the calculations lie within a boundary.
So imagine a few hundred calculations, all acting on the same 6 inputs with their varying values, where 90% of the calculations are within 'close proximity' to each other. (Try to picture a graph with a box lying on the graph where 90% of the 'dots' lie within the box)
I don't think that the method I am after is the Monte Carlo method (where if I remember correctly, just replicates the calculations randomly over a subset to re-run simulations). Or is it? Or is it some other sampling/statistical methodology?
Apologies if this is still a bit vague but I am stumped at the moment
Many thanks
Many, many moons ago when I was at university, one of my close friends was studying statistics and of course we talked about or respective subjects.
One of the things that he mentioned was a method allowing probability/statistics to help produce a set of variables to try and predict an outcome.
Probably not explaining this well but I now need that sort of algorithm at work but I cannot remember the name of the methodology.
Can someone help?
Basically I have a multi-variable input scheme (about 6 inputs) that vary in value from 0 to an upper limit (each independent and known).
I want to be able to analyse these variables over their input range to produce an output set where 90% of the results of the calculations lie within a boundary.
So imagine a few hundred calculations, all acting on the same 6 inputs with their varying values, where 90% of the calculations are within 'close proximity' to each other. (Try to picture a graph with a box lying on the graph where 90% of the 'dots' lie within the box)
I don't think that the method I am after is the Monte Carlo method (where if I remember correctly, just replicates the calculations randomly over a subset to re-run simulations). Or is it? Or is it some other sampling/statistical methodology?
Apologies if this is still a bit vague but I am stumped at the moment
Many thanks
Answers
1 to 6 of 6
Best Answer
No best answer has yet been selected by GSD4ME. Once a best answer has been selected, it will be shown here.
For more on marking an answer as the "Best Answer", please visit our FAQ.
Yes certainly sounds like multiple regression analysis is what you’re talking about.
This involves looking at variations in the independent variables (your six “inputs”) and analysing their influence on the dependent variable (the data you are trying to forecast). It is a fine method except that in trying to forecast the dependent variable you need forecasts of the independent variables and these may be much more difficult to forecast!
Six independent variables would certainly involve a powerful “engine” to perform the analysis and the relationship between the six and the dependent variable (including your certainty about a “causal” relationship, which you must have if the analysis is to be valid) may be difficult to summarise. There are a number of statistical tools to help you validate the relationship, but it would still involve some possibly shaky assumptions.
In my earlier life I was involved in forecasting using MRA and we tried to restrict the independent variables to three at most, together with some “step” changes ( variations not explained by the independent variables) and seasonality. The reason for this was that as the number of IVs increase the reliability of the model decreases - sometimes alarmingly. In addition, as I mentioned, you also have the problem of forecasting all of the IVs if you want to use the model to produce a forecast.
This involves looking at variations in the independent variables (your six “inputs”) and analysing their influence on the dependent variable (the data you are trying to forecast). It is a fine method except that in trying to forecast the dependent variable you need forecasts of the independent variables and these may be much more difficult to forecast!
Six independent variables would certainly involve a powerful “engine” to perform the analysis and the relationship between the six and the dependent variable (including your certainty about a “causal” relationship, which you must have if the analysis is to be valid) may be difficult to summarise. There are a number of statistical tools to help you validate the relationship, but it would still involve some possibly shaky assumptions.
In my earlier life I was involved in forecasting using MRA and we tried to restrict the independent variables to three at most, together with some “step” changes ( variations not explained by the independent variables) and seasonality. The reason for this was that as the number of IVs increase the reliability of the model decreases - sometimes alarmingly. In addition, as I mentioned, you also have the problem of forecasting all of the IVs if you want to use the model to produce a forecast.
1 to 6 of 6
Related Questions
Sorry, we can't find any related questions. Try using the search bar at the top of the page to search for some keywords, or choose a topic and submit your own question.