As stated above, historically the specified
maintainability index is also the one that is demonstrated. However, it is
important to provide some guidance on choosing such an index, as this can
affect which of the test methods, outlined in Table B-I , is chosen for Maintainability Demonstration testing.

The principal criterion in selecting the index for a maintainability demonstration (and therefore for a product specification) is consistency with the mission objectives and operational constraints. This will generally mean that equipment downtime is the time measurement of the index since operational effectiveness cannot be achieved unless downtime is controlled. If the need for an equipment or system is not critical, and manpower control is important, a labor-hour index may be most appropriate. Preventive-maintenance labor-hours per operating hour is preferable to downtime due to preventive maintenance for equipments for which such required maintenance can be scheduled without fear of operational demand during the maintenance action.

By the same reasoning, corrective maintenance is more crucial than preventive maintenance, especially if the latter can be scheduled to take place during known periods of non-use. For equipment operated or needed continuously, such as an alert radar, total maintenance time is of prime importance. For equipment demanded at random times, such as a missile-defense system, the approach might be to use separate controls for corrective maintenance and preventive maintenance. The choice of the statistical measure to be used often depends on the mission objective. If there is an operational availability requirement for the system, then the equation for A_{o}, reprinted below from Section
2.2.1.2 of the handbook, is used to determine maintainability requirements.

Inherent availability may also be a requirement, in which case MTBF and MTTR are substituted for MTBM and MDT, respectively, in the above equation. When either availability expression is appropriate, then a mean value becomes the maintainability index to be demonstrated. There may be, however, an availability requirement where a maximum downtime or mean time to repair is required or more appropriate. Such a requirement would apply to critical equipment aboard an aircraft, for example, where the aircraft may have to be available for a new mission within two hours after completing a mission. In this case, a requirement of 0.95 probability, for instance, of completing the necessary maintenance within 100 minutes would be more consistent with the operational objective than a mean-value index. Of course, the maintenance level for which the requirement is developed also plays a role. For instance, a maximum time to repair may not be appropriate or needed at the intermediate or depot-level of maintenance assuming that an adequate amount of spares are available at the next lower level of maintenance (i.e., organizational or intermediate, respectively).

To provide further guidance in choosing an appropriate index for demonstration, a guidance matrix is provided in here in Figure B-2.^{2} To use the matrix, each of the conditions listed at the top of the exhibit that apply to the equipment of interest should be checked. The appropriate index is then found from the matrix by locating the column that contains an x for each condition checked above. For example, if steady-state availability is a critical parameter (Condition 1) and maintenance time is limited by environmental or operational circumstances (Condition 5), the recommended index provides a control on both the mean and maximum maintenance time, and there is an option for including preventive-maintenance time depending on equipment use or scheduling and criticality. The set of conditions listed is not exhaustive, but it is believed to include the most important ones.

Figure B-2 - Procedure For Maintainability-Index
Selection

Several of the major considerations that led to the development of the matrix are as follows:

- The mean is directly related to steady-state
availability and is therefore the index of choice when this operational
requirement exists.
- If the distribution of maintenance times is
unknown, the median is preferred since it permits distribution-free tests.
If availability is critical, however, use of the central-limit theorem
permits a mean test, provided the sample size is large.
- For the lognormal distribution, the median is
preferred to the mean (assuming that Condition 2 applies and that 5 and 6 do
not) since it is based on only one parameter, which makes statistical
analysis exact.
- When maintenance time is limited (Condition 5), the M
_{Max} index is preferred.
- The mean is preferred over the median if manpower control is also required because the mean is more directly related to manhours. However, if the distribution is unknown, the median may be used as long as availability is not critical.

Note that complete dependence on this procedure is to be avoided. Because of the wide variety of equipments, mission objectives, and environmental and operational circumstances, the selection matrix should be considered a guide only. Ultimately, the best measure is determined by individual system circumstances and good judgment.

Finally, no matter what indices are specified in the requirements or as the index to be demonstrated, the values for those indices must be realistic, based on current knowledge of the state-of-the-art, past history of like systems, and engineering judgment. Whether historical data or prediction, or both, is used for assessing realism, careful judgment is required. If an allocation leads to an M_{ct} value of 20 minutes, but a 30 minute value was observed for the most similar existing item, can it be concluded that 20 minutes is achievable? In all such cases, the following questions should be considered:

- How similar are the items?
- How similar will the maintenance environment be?
- Since the observed 30-minute value is necessarily based on a sample, what is the lower confidence limit associated with such a mean-value estimate?
- How much maintainability improvement can reasonably be expected?
- Is there any margin for increasing the 20-minute specified value?

^{2}Taken from RADC-TR-69-356, "Maintainability Prediction and Demonstration Techniques, Volume II - Maintainability Demonstration," Technical Report, ARINC Research Corporation, January
1970.