Home / Monte Carlo / Analyzing the Model / Percent Contribution
Percent Contribution¶
From Excel click...
QXL Monte Carlo > Monte Carlo > Contribution Tools > Percent Contribution...
Decomposes the variation of each chosen output across the model's inputs, so the report says how much of an output's variation each input accounts for.
Three methods are offered, and you may run any combination of them. The dialog lists each with its own description; see Options.
One simulation per method¶
Each ticked method gets its own simulation. Asking for all three is three simulations, and LMG dominates the time.
All three are given the same random seed, so wherever two methods use the same number of simulations they are computed on identical trials, and the comparison between the methods is not confounded by different random numbers.
Fast Mode and Native Mode¶
A Percent Contribution run goes through the same path every simulation goes through, including the check that proves the engine against Excel before it is trusted.
The contribution table is produced in both modes. When the model runs in Native Mode, where Excel evaluates each simulation, the report is still written. A Native Mode run is limited to 100,000 simulations, so a method set to more than that runs 100,000.
Validate Model names what forces Excel: the formulas the engine refused, or the output where the engine's result disagreed with Excel.
What it needs¶
At least one marked input and at least one marked output that collects statistics. A workbook missing either is refused before the dialog opens. With no marked inputs the message is:
This workbook has no marked inputs. Percent Contribution splits an output's variation among the inputs that drive it, so at least one input must be marked.
With no marked output that collects statistics:
This workbook has no marked outputs that collect statistics. There is nothing to decompose.
The report¶
Every ticked method is written to one worksheet, Percent Contribution.
The block at the top left lists the Process Inputs, one line per method with its number of
simulations (for example LMG: 10,000 Trials), and the Random Seed every method used.
To the right, each output has its own column band, starting at column G, with a second output at column P, and so on. Down each band there is one block per analysis, in the dialog's order:
| Block heading | From |
|---|---|
| output: LMG Shapley Regression (Dominance Analysis) | LMG |
| output: OLS Type I (Sequential Sums of Squares) | Least Squares |
| output: OLS Type III (Adjusted Sums of Squares) | Least Squares |
| output: Spearman Rank Correlation | Spearman's Method |
Least Squares gives two blocks because the one fit produces both decompositions.
Each block is a heading, a statistics line, a Pareto chart of horizontal bars sorted longest
first, and the chart's numbers in a table directly under it. The statistics line gives
Simulations and Invalid Simulations for that method and that output, then R-Squared where the
method has one, and for the two Least Squares blocks Total Sum of Squares and
Residual Sum of Squares.
The table columns are Input and % Contribution Variation, with a Sum of Squares column between
them in the two Least Squares blocks. After the inputs come the remainder rows, set off in gray
italics:
| Row | Blocks | Value |
|---|---|---|
| Shared Variation | OLS Type III only | R-Squared minus the sum of the Type III contributions. It can be negative, and is then drawn as a bar to the left of zero |
| Unexplained Variation | LMG, OLS Type I, OLS Type III | 1 minus R-Squared |
The Spearman block has no R-Squared and no remainder rows; its contributions total 100 percent.
A Type III block whose Shared Variation is below zero carries a highlighted note giving the two figures. Each block ends with a paragraph about its method, and the sheet ends with a paragraph beginning "Note on interpretation:".
The report does not say whether an input raises or lowers an output.
Related¶
- Options, for the three methods, the simulation counts and the seed.
- Run Model, the ordinary simulation.
- Validate Model, for what makes a model run in Native Mode.