First Part

The Uncertainty of the Past: Organizational Learning Under Ambiguity

Learning From Experience

How does learning from trial and error proceed?

Feedback from previous experience is used to select among alternatives. In this case learning happens from experimentation, evaluation, and assessment.

But we know that rationality is limited in:

* awareness of alternatives

* precision/interpretation of information

* clarity and consistency of goals

What happens in the experiment is unclear, and the goals used in the assement are ambigous too.

What is the complete learning cycle?

1. Individual actions or participation in choice situation

2. Organizational actions, choices, outcomes (via cum indiv. action)

3. Environmental response or actions

4. Individual cognitions, interpretations, and "models of the world".

How is experience interpreted?

Experience is interpreted differently by everyone. People try to be rational in their interpretations, but they are confounded by ambiguities and awareness limits. Reality is also processed through a filter of past experiences and biases, and is also constrained by expectations.

Interpretation is also situational. The fundamental attribution error also is in effect (as well as self-ego reinforcements).

Individuals also probably have a success/failure dichotomy in interpretation.

How are the lessons of experience stored and implemented in organizations?

Primarily via rules, though also with rumors, legends, myths. They also are embued in the roles that people take.

What are aspiration levels?

Goals that people try to attain.

What is satisficing?

Satisficing is ending your search when the first acceptable alternative is presented. It most often occurs when all alternatives are unknown or it costs to much to evaluate them all.

In what sense is satisficing a theory of search and attention, rather than a theory of utility functions?

With utility functions one has an understanding of the entire range of possibilities, and has developed preferences for each. In this case a person could then move toward the optimal alternative with little difficulty.

In satisficing the person is not aware of all the potential alternatives nor is concerned with optimization. They have a goal in mind and will evaluate those alternatives that come to their attention. In their search they only evaluate those alternatives that are discovered or brought to their attention. They are also more likely initially give attention to those alternatives that are easier to evaluate and then move on to more difficult ones. Often there is not system to their search, and it is rarely exhaustive.

What is the relation between satisficing, aspiration levels and learning from experience?

All are results of applying bounded rationality to decision making.

Incomplete Learning Cycles

What are the major forms of incomplete learning cycles?

What difference to they make?

1. Role-constrained learning

Individual learning has no impact on behavior. People do what is expected of their role or standard operating procedures take over. I guess that preconceived biases also has an effect too.

Difference: It means that despite the best information one cannot get people to act (or change their actions). Getting employees to live a more healthy lifestyle to save on medical benefits is one example.

2. Superstitous Experiential Learning

In this situation organizational actions have little impact on environmental response, so learning occurs unreliably. People have to rely on trust, faith, or superstitions to intepret results of actions (because they are not directly observable).

Difference: People may learn the wrong things because they can't observe the impact of their actions, or they conclude their actions have no impact. Pollution or "tradegy of the commons" is one example.

3. Audience Experiential Learning

Here individual action has little impact on organizational action. Learning occurs but not organizatonal action. Learning in politics or research happen this way.

Difference: Here individual action doesn't imply collective action. Janus's groupthink is one where almost everyone is opposed to an action but silence implies approval to others which eventually produces an agreement to procede.

4. Learning under ambiguity

Here it is hard to tell what happened or why. Causal connections must be inferred. Records are not completely accurate or comprehensive. Variation in organizational memory affects organizational beliefs. Incentives can affect the interpretation of success and failure.

Believes are sensitive to the timing, order, and context of information.

Difference: People may infer the wrong things from their actions. Learning may procede away from improvement. Disagreements may arupt about intrepretation of actions. Different interpretations may result in different actions.

In what ways would a model of "seeing liking and trusting" contribute to understanding learning in organizations?

It shows how roles, expectations, and attitudes affect how people see and interpret the world and thus how organizations learn.

The roles and cultures set up in organizations can affect how the organization perceives it's environment and how the environment reacts to its actions. Expectations can create self-fulling prophecies. Roles can limit flexibility of action, especially in times of adversity or emergency. Organizational histories create organizational beliefs that affect cognition and interpretation.

Objective reality, attitude structures, social reality, and social norms thus impact on behavior.

The model shows how integration and alientation are self-reinforcing. People preferentially prefer information they trust. They "emphasize the impact of interpersonal connections in organizations and the affective connection between the organization and participant on the development of beliefs.

What are some problems?

In conditions of ambiquity, deciding what you expect to see or relying on people you trust can be problematic. In rapidly changing organizations the social reality may be evolving too fast to interpret. Newcomers to organizations may react differently.

Unfortunately, this process does not guarantee that learning will result in improvements over time. Ambiquity in evaluation processes can severely hamper effective learning.

Second Part:

A Model of Adaptive Organizational Search (Levinthal & March)

What is the Basic Structure of the Model?

Two Search Methods : Refinement (R) and Innovation (I)

* success favors more innovation search

* failure favors more refinement search

The ability (resources) to search (U) depends on prior success or failure:

* success means more resources for innovation search

* failure means more resources for refinement search

Search efficiency is improved through experience with current product technology (E)

Search propensity depends on sucess/failure in the previous time period (S)

* success increases propensity for innovation search

* failure increases propensity for refinement search

Research funds for search depend on available resources (U) and search propensity (S)

Technology search is modeled by drawing samples from a "refinement pool" and an "innovation pool". The variance of the "refinement pool" is high after an innovation but decays over time. The variance of the "innovation pool" is a function of the existing technology, which means organiztions with better technology are potentially able to find bigger innovations. There is also a estimation error added to each sample to represent search ambiguity.

After sampling from both pools the organization chooses the highest value technology to represent its result. Performance is the value of this technology times a random uncertainty factor minus search costs.

The resulting performance is then compared versus the goal for that time period to determine success or failure. The goal for the next period may then be modified based on changing aspirations.

Some control variables in model:

* environmental uncertainty (affect performance each time period)

* current technology value may increase, decrease, or remain the same over time

* search estimation error (org could find a winner but not know it or change to a loser)

* goal adaptation rate (how change bar with success)

What is the difference between refinement search and innovation search and how are they encompassed in the model?

Refinement search has a narrower "pool distribution", meaning less chance for signficant success or failure. It is triggered and reinforced after each failure. However, the variance of results decays over time, meaning that the potential for major technology improvements via refinment is reduced with time. The more an organization focuses on refinement the less chance they will get an innovation.

In refinement mode search efficiency will increase but potential returns decrease (without an innovation). In innovation mode each innovation increases the potential of even bigger innovations.

What Are the Four Major Clusters of Results and why might they be significant to understanding organizational learning?

1. Sensibility

Organizational learning results from a sensible learning process in a confusing world. Overall, the model organizations tended to improve over time.

Time horizon impacted the final results. For short-term gain, minimal search investment was the best strategy. For longer horizons, full investment was the best strategy. Most organizations learned to spend 10-40% on search.

A large search propensity added risk to the organization, resulting in higher variation of performance.

Note that:

* reward for refinement search decays with refinement draws

* efficiency of search step declines with each innovation

* when technical opportunities are such that it makes finding an important technology likely, learning increases propensity (through success).

2. Success

Most organizations were sucessful.

Short run changes in resource allocation was less responsive to learning than to organizational success or failure.

In improving environments organizations spend more on innovation than refinement, & as a result get more efficient at it.

A tendency to failure produces more spending on refinement.

High environment uncertainty reduces frequency of subjective success when technology is improving or declining slightly, but reduces frequency of failure when tech is declining signficantly (i.e., one must innovate to survuve).

3. Path Dependence

Draws occaisionally yield extreme values. A low extreme are somewhat irrelevant (dont pursue them) but high value permanently changes the organizational position).

One random value (e.g., a big change) affects the probablity of the next step.

"Organizational histories are produced through a combination of chance events and adaptations to those events that, in some cases, considerably amplify the effects of chance".

A success leads to success, a failure leads to failure.

A success: * increases organizational slack

* increases expenditures on innovation and propensity to innovate

Failure is:

* sensitive to rate of adapting goal to performance

* degree of environmental uncertainty

* size of external changes to technology

4. Sensitivity to Learning Rates

For long horizon, set search rates as high as possible. However, the strategy is chancy and you often need multiple time periods to see a return.

Each organization is liable to find a spectacular technology and have a long string of sucesses.

Fast learners adapt quickly to correct and false signals, which can lead to persistent mistakes (jump to refinement too quickly).

Long run performance improved through relatively slow, imprecise learning (don't shift search strategies so fast).

Performance is highest when the short run doesn't indicate searching or when the average quality of outcomes increases with experience (you can count on the learning).

If one reacts too quickly they lower search propensities too fast, which reduces expenditures and lower chances of finding a new technology.

But after 30 periods fast learners are better than slow learners. Still, rapidly changing goals makes success more problematic.

Summary

Model show why we have problems in finding consistent factors with organizational change and search. The model shows the power of being lucky (if you invest in luck producing search routines).

How might the model be elaborated in a useful way?

Often companies need to lauch a product before they find out if it is successful rather than know in advance and choose among various alternatives (current, new, etc.). Could we force companies in the model to "live" with a new poor technology for a while or charge a cost to return to the old technology?

The value of a new improved technology is not realized instantaneously. Could you ramp up the performance over a few time periods instead to represent the uncertainty involved during the lauch process?

It is uncertain whether potential returns to an innovation search is necessarly lognormal. One would think the really bad end would be eliminated with improved search efficiency. Can we modify the distribution?

One could also add in the risk aversion near the survival point.