Skip to content
Back to search results

Fisher-Pry Substitution Model

Introduced by
John C. Fisher & Robert H. Pry (General Electric)
Year
1971
Type
Logistic-growth forecasting pipeline for technological substitution
Family
Trend extrapolation / substitution curves
Font size

What does it do?

The Fisher-Pry Substitution Model, introduced by John Fisher and Robert Pry of General Electric in 1971, is a foundational technique for forecasting technological substitution — the process by which a new technology displaces an older one.

Its empirical premise, supported by dozens of historical cases from synthetic to natural materials and from one manufacturing process to another, is that the fractional market share captured by the new technology follows a logistic (S-shaped) growth curve. The model's elegance lies in a transformation: when the takeover ratio f/(1-f), the ratio of the new technology's share to the old's, is plotted on a logarithmic scale against time, the substitution traces a straight line. This linearization makes it easy to fit, interpret, and extrapolate substitutions from sparse early data, which is why the Fisher-Pry curve remains a workhorse of technological forecasting.

The intuition behind the method

New technologies rarely take over all at once; they creep in, then surge, then saturate, producing the familiar S-curve of adoption. Fisher and Pry's key observation is that the two assumptions behind such substitutions — that once it begins, substitution proceeds to completion, and that the rate of takeover is proportional to both how much has been taken over and how much remains — are exactly the assumptions that generate a logistic curve. The slow start reflects the small initial share; the steep middle reflects the moment when both adopters and remaining holdouts are plentiful; the slow finish reflects the dwindling pool left to convert. The clever part is the change of variables: instead of the awkward S-curve, you track the ratio of new to old, f/(1-f). On a logarithmic axis this ratio plots as a straight line, so forecasting becomes as simple as fitting a line to early points and extending it. The line's slope tells you how fast the substitution is happening, and the model even pins down a handy summary — the years it takes to go from ten percent to ninety percent takeover.

History

John C. Fisher and Robert H. Pry, both at General Electric, published 'A simple substitution model of technological change' in Technological Forecasting and Social Change in 1971. Examining numerous historical cases — synthetic versus natural fibers and rubber, plastics versus traditional materials, and many manufacturing processes — they found that the fractional takeover of a new technology consistently followed a logistic curve, and that the odds ratio f/(1-f) plotted as a straight line on semilog paper. Their paper distilled this into a two-parameter model that could be fit by hand from sparse data, and it became one of the most widely used and cited tools in technological forecasting. Later work by Marchetti, Nakicenovic, and others generalized it to multiple competing technologies and to broader logistic-systems analysis, but the Fisher-Pry transformation remains the canonical entry point.

When to use it

Use the Fisher-Pry model when one technology, product, material, or process is clearly displacing another in a shared market or function, and you have a time series of the new alternative's fractional share that already shows the substitution underway. It is ideal for forecasting how quickly the takeover will complete and when key share thresholds will be crossed, especially when early data are limited but a logistic mechanism is plausible. It is poorly suited to cases where multiple technologies compete simultaneously (where multi-mode substitution models are needed), where the new technology may not go to completion, where the market itself is the thing growing rather than a share being transferred, or very early before any substitution trend is visible. It also presumes the displaced technology is not itself improving fast enough to reverse the trend.

Limitations and considerations

  • Assumes substitution proceeds to completion, which fails when the incumbent survives or rebounds.
  • Handles only one-to-one substitution; simultaneous multi-technology competition needs richer models.
  • Forecasts are fragile in the early phase when few points constrain the slope and midpoint.
  • The constant-rate logistic shape ignores shocks, policy shifts, and improvements in the old technology.

Source conditions do not replace independent scientific review or validation by an analysis engine.

Strengths

  • Reduces S-shaped substitution to a straight line, making fitting, diagnosis, and extrapolation simple and transparent.
  • Requires only a fractional-share time series and yields interpretable parameters: a rate and a midpoint.
  • Delivers a single intuitive summary of speed — the ten-to-ninety-percent takeover time ln(81)/alpha.
  • Validated across many historical technological substitutions, giving it strong empirical pedigree.

Applications

  • Forecasting how fast a new material or process will displace an incumbent in industry.
  • Projecting market-share transitions between competing consumer technologies.
  • Estimating when an emerging technology will cross key adoption thresholds.
  • Providing baseline substitution trends as inputs to trend-impact and scenario analyses.
  • Retrospectively characterizing the speed of historical technological transitions.

How it works

How it works

1

The pipeline begins by defining the substitution fraction f(t): the share of the relevant market, output, or installed base held by the new technology relative to the combined total of new and old. Fisher and Pry frame substitution as a competition between two alternatives serving the same function, so f runs from just above zero, when the newcomer first appears, toward one, when it has fully displaced the incumbent. Getting this fraction right — a clean, consistently measured ratio of new to (new plus old) over time — is the essential data step, because the entire model is built on the dynamics of this single fractional series.

Formula
f(t) = \frac{N_{\text{new}}(t)}{N_{\text{new}}(t) + N_{\text{old}}(t)},\quad 0 < f < 1
2

The model then imposes its core behavioral assumption: that the rate of substitution is proportional to the product of the fraction already taken over and the fraction remaining, df/dt = alpha f (1 - f). This is the logistic differential equation, embodying Fisher and Pry's two postulates — that substitution, once underway, tends to completion, and that its speed peaks in the middle. Integrating this equation yields the closed form in which the odds ratio f/(1 - f) grows exponentially in time, equal to exp(alpha (t - t0)). The single rate parameter alpha governs how quickly the takeover proceeds, and t0 marks the midpoint at which the new technology reaches fifty percent share.

Formula
\frac{df}{dt} = \alpha\, f\,(1 - f)\ \Longrightarrow\ \frac{f}{1-f} = \exp\!\bigl(\alpha (t - t_0)\bigr)
3

The decisive analytical move is linearization. Taking the natural logarithm of the odds ratio turns the exponential into a straight line: ln[f/(1 - f)] = alpha (t - t0), a linear function of time with slope alpha and a zero crossing at the midpoint t0. Equivalently, plotting f/(1 - f) on a logarithmic (semilog) vertical axis against time on a linear horizontal axis renders the substitution as a straight line. This transformation is the heart of the Fisher-Pry method: it converts a nonlinear S-curve into a form where the eye, or a simple regression, can immediately judge fit, read off the rate, and extrapolate.

Formula
\ln\!\left(\frac{f}{1-f}\right) = \alpha (t - t_0) = \alpha t - \alpha t_0
4

With the relationship linearized, the analyst estimates its two parameters from the observed data. The slope alpha and the midpoint t0 (equivalently the intercept) are obtained by fitting a straight line to the points ln[f_i/(1 - f_i)] versus t_i, typically by least squares. Because the model is linear in the transformed variable, this is an ordinary linear regression, which also yields a visual and statistical check on how well the logistic assumption holds for the case at hand. A good straight-line fit over the observed period is the evidence that the Fisher-Pry mechanism is operating and that extrapolation is warranted.

Formula
\hat\alpha,\ \hat t_0 = \arg\min_{\alpha, t_0} \sum_i \Bigl[\ln\tfrac{f_i}{1-f_i} - \alpha (t_i - t_0)\Bigr]^2
5

Finally the fitted parameters are used to forecast the substitution. Inverting the transformation gives the projected share at any future time, f-hat(t) = 1 / (1 + exp(-alpha-hat (t - t0-hat))), the logistic curve itself, from which one reads when the new technology will reach any target share. A particularly useful summary the model delivers is the takeover time from ten to ninety percent, equal to ln(81)/alpha-hat, a single number characterizing the speed of the transition. These projections, drawn from sometimes sparse early data, are the practical payoff: anticipating how fast and how far a new technology will displace the incumbent.

Formula
\hat f(t) = \frac{1}{1 + \exp\!\bigl(-\hat\alpha (t - \hat t_0)\bigr)},\quad \Delta t_{10\to90} = \frac{\ln 81}{\hat\alpha}

Common pitfalls

  • Extrapolating from too few early data points, where the fitted slope is highly uncertain.
  • Applying the model when the market is expanding rather than a fixed share being transferred.
  • Ignoring that the displaced technology may improve and halt or reverse the substitution.
  • Forcing a single logistic on data that actually reflect several overlapping substitutions.

Frequently asked questions

What do the parameters alpha and t0 mean?

Alpha is the substitution rate: the slope of the straight line on the semilog plot, governing how fast the new technology takes over. The midpoint t0 is the time at which the new technology reaches fifty percent share, where ln[f/(1-f)] crosses zero. A convenient derived quantity is the ten-to-ninety-percent takeover time, ln(81)/alpha, which expresses the whole transition's speed in a single number of years. Together alpha and t0 fully specify the logistic curve and hence the entire forecast.

Why transform to f/(1-f) instead of fitting the S-curve directly?

Because the transformation linearizes the problem. Plotting the odds ratio f/(1-f) on a logarithmic axis against time turns the logistic S-curve into a straight line, so you can fit it with ordinary linear regression, judge goodness of fit by eye, and extrapolate trivially. The straightness of the transformed data is also a built-in diagnostic: if the points lie on a line, the Fisher-Pry mechanism is operating; if they curve, the simple model is suspect and a richer one may be needed.

When does the Fisher-Pry model break down?

It breaks down whenever its assumptions are violated. If the substitution does not go to completion — because the incumbent improves, regulation intervenes, or the new technology stalls — the logistic shape will mislead. If several technologies compete at once, a single two-way substitution model is inappropriate and multi-mode logistic methods are required. And in the very early phase, with only a few data points, the estimated slope and midpoint are highly uncertain, so forecasts should be treated as provisional until the trend is well established.

Related methods

Sources

  1. Fisher, J. C., & Pry, R. H. (1971). A simple substitution model of technological change. Technological Forecasting and Social Change, 3, 75-88.

    Open DOI reference
    Copied.