---
id:
aliases: []
tags:
  - destiny/uncertain
  - status/draft
  - topic/estimating
  - type/philosophy
title: Estimating Methodologies
---
# Estimating Methodologies

## The 4 Traditional Methodologies

> [!quote] Defense Acquisition University
> There are four principal cost estimating methodologies:
> 1. Comparison/analogy,
> 2. Parametric,
> 3. Detailed engineering/bottom up, and
> 4. Extrapolation from actual costs.

### Comparison / Analogy

**Pros:**
* simple to implement
* quick in practice

**Cons:**
* unsatisfying input
* unsatisfying output

### Parametric

**Pros:**
* quick in practice
* nuanced input
* accounts for incomplete input
* effectively accurate through the entire estimate process
* well-suited for automation

**Cons:**
* complex to implement

### Detailed Engineering / "Bottom-Up"

**Pros:**
* simple to implement
* satisfying output

**Cons:**
* slow in practice
* unsatisfying input
* accuracy and risk depend solely on estimator
* output dependent on complete input
* ill-suited to automation

### Extrapolation from Actual Costs

Depending on the application this is one is either
* better described as the previous 3
* or not estimating at all.

## In Other Industries

There exist articles from software development project management resources
stating essentially what I believe to be true,
that parametric estimating is _more_ accurate than bottom-up,
in contrast to common assumption.

I suspect that industry tolerance for cost-modeling
is related to the cost of estimation.
Software is largely impractical to estimate bottom-up,
whereas anyone can take off construction scope.

Alternate theory:
tolerance is related to the industry's ability to understand the model.

Regardless, a continuum emerges:

```
^
| Software
| Defense
| Construction
v
```

Industries that would benefit the most from cost-modeling
are least likely to consider it a valid method.

These methods are actually the same, just different levels of abstraction.
This can be observed in the _very_ subtle distinction
between parametric and analogous estimating.

```
# of Parameters
^
| Bottom-Up     (100+)
| Parametric     (10)
| Analogous       (1)
| Extrapolation   (0)
```

These different methods are just a response
to different levels of requirement specificity.
Parametric is obviously more accurate and precise than analogous,
however the question not being asked
is if there are diminishing returns to estimate specificity.

Suppose there are: at some point then,
there must then be a point at which **risk of human error**
outweighs the value of more perfect information.