Short-Range Actuarial Projections of the Old-Age, Survivors, and Disability Insurance Program, 2001
Actuarial Study No. 115
Chris Motsiopoulos and Tim Zayatz, A.S.A.
Contents Previous Next Glossary




Each month, benefits are paid to individuals in current-payment status as of the end of the previous month. Adjustments are then made for benefits awarded and terminated during the current month. Finally, adjustments may be made for items such as benefit recalculations for additional earnings, and annual cost-of-living increases. This section describes the estimation of the average amount of new awards; the other factors involved in estimating benefit payments will be discussed in later sections.

1. Sample Data

Projecting the average amount of a new award involves an actual sample of earnings histories, drawn from the Continuous Work History Sample (CWHS). The CWHS is a 1-percent sample of all people with earnings covered by Social Security. The sample drawn from the CWHS consists of 10 percent of all persons who were not entitled to an OASDI benefit as of the end of 1987, but were entitled to a benefit-either as a retired worker, disabled worker, or survivor-by the end of 1988. Thus the sample is a 0.1-percent sample of workers with benefits newly entitled as of calendar year 1988.

Dependents of retired and disabled workers are not included in the sample; the model assumes these types of benefits are proportional to the primary benefit. Beneficiary information from the CWHS includes the Primary Insurance Amount (PIA) and monthly benefit as of December 1988, the worker's gender and date of birth, any period of disability, and other benefit information. The sample includes records on 2,229 beneficiaries, distributed as shown in the "newly entitled" column of the following table:

Type of beneficiary
  Actual awards
(in thousands)
  Newly entitled
(from sample)
Male old-age beneficiary
Female old-age beneficiary
Young survivor
Aged survivor
Disabled survivor
Male disabled worker
Female disabled worker

As you can see, the time of entitlement is not necessarily the same as the time of award, as "entitlement" and "award" have different meanings (see Glossary). The number of newly entitled beneficiaries is generally less than the actual number of awards for several reasons. First, entitlement can be retroactive and thus may precede the date of award. For example, an individual entitled in 1988 may actually be awarded in a subsequent year; or an individual awarded in 1988, may have been entitled in a prior year. This effect is especially apparent for disabled workers, where the disability determination process might result in an award months or even years after entitlement.

Second, the newly entitled are those workers whose family composition generates awards in the various categories; whereas the number of actual awards includes these auxiliary beneficiaries. For example, one newly deceased worker with two eligible children and a dependent mother would be counted as one newly entitled, but would result in three additional awards.

2. Sample Adjustments

The next step in projecting the average award amount is to construct a simulated sample of awards for the year prior to the projection period-2000 for this study-and for each year in the projection period-2001-10. Average amounts are then constructed for each sample. The series of average amounts is then converted to an index representing benefit growth, having a value of 1.000 in 2000. The index for each projected year is then applied to the actual average benefit awarded in 2000-as determined from actual Social Security data-to obtain the final projection of award amounts.

The simulated samples are biased for several reasons. First, they reflect the composition of OASDI beneficiaries in 1988, which may not persist in the future. In particular, the fraction of all females with enough earnings to be eligible for retirement benefits at normal retirement age or earlier was 68 percent in 1988; currently it is 76 percent, and is expected to increase to 84 percent by 2010. Females who were not insured for old-age benefits in 1988 could not have been drawn from the CWHS in the initial sample selection. Therefore, an adjustment is needed to properly represent them in the simulated samples for projected years. Earnings records of some uninsured females in the constructed samples are modified to ensure that a reasonable percentage of female workers become insured for old-age benefits in the future.

Another reason for sample bias is that samples were not drawn from the population as a whole. Rather, they were taken from the CWHS, which is itself a sample. As an adjustment, two types of uninsured workers were added to the sample drawn from the CWHS to make it representative of the full population:

A total of 300 records-80 male and 220 female-of the first type were created and added to the sample. These records are based on data drawn from a complete population survey with earnings matching Social Security records. A total of 100 records-all female-of the second type were also created and added to the sample.

Note that no adjustments were made to the number of disability and survivor records in the sample because of the difficulty in defining the appropriate population. All-in-all, the adjusted sample for 1988 includes a total of 1,947 records of retired workers-999 male and 948 female-in addition to the original disability and survivor beneficiary records.

3. Sample Earnings

Each simulated sample for 2000-10 contains potentially the same number of beneficiaries as the adjusted 1988 sample, along with the same age and gender characteristics. To account for future employment, the model updates the earnings record of each worker in each sample to create representative earnings records that would underlie an award in a particular year. The steps involved in creating simulated sample earnings is described below.

The benefit computation procedures that generally apply consider earnings after 1950, up to the year of award. So as the year of award shifts from 1988 through 2010, the length of the earnings record for each worker increases. For example, a 65-year-old retiring in 1988 had annual earnings from 1951 to 1988 in the original (adjusted) sample-this represents earnings for ages 28 to 65-while prior earnings were aggregated for pre-1951. The "parallel" 65-year-old retiring in 2001 is assigned annual earnings from 1951 to 2001 in the simulated sample for 2001-this represents earnings for ages 15 to 65. To create an earnings record while retaining the characteristics of the worker in the original 1988 sample, year-by-year earnings are expanded by duplicating some randomly-selected years. Since the exact pattern of earnings near the time of retirement is crucial, these years are not considered in the random selection. Note that for disability and survivor cases, earnings for workers under age 55 in 1988 are not expanded; in general, these workers had a full earnings record by 1988 with little or no pre-1951 earnings.

The level of annual earnings in each simulated sample is based on the original 1988 sample, adjusted to reflect annual wage increases. Another adjustment to earnings is required because the 1988 sample includes earnings only up to the contribution and benefit base. As a result of ad hoc increases in the wage base for 1979-81 and the automatic-adjustment mechanism, the wage base will be higher during the projection period-relative to average earnings-than during the years prior to 1979. To approximate the amount of annual earnings in excess of the wage base for each record in the 1988 sample, a random selection procedure was used with a statistical distribution of earnings in excess of the wage base, as provided by the Office of Research, Evaluation and Statistics.

Further adjustments to annual earnings are made in the simulated samples to reflect changes in:

Females who originally had some years of zero earnings are randomly selected to have positive earnings to increase rates of participation in covered employment, and produce fully-insured status in the simulated samples. In addition, earnings for females may be increased and earnings for males may be slightly decreased to narrow gender differences.

4. Earnings Prior to 1951

Finally, adjustments are made to account for the fact that year-by-year earnings are not available prior to 1951. A 62-year-old retiring in 1988 was 25 in 1951, so the expansion of earnings records from the original 1988 sample is done on the basis of earnings beginning at age 25. In general, earnings are lower at the younger ages, even after considering wage increases; therefore, adjustments to projected earnings at younger ages are needed to assure that overall coverage rates were reasonable. Average earnings for both males and females are generated so that annual wage increases over the projection period closely match assumed rates of increase.

Some benefit calculations use earnings prior to 1951, which are usually reported as a lump-sum total. These earnings continue to be applied in the future-up to a point-and must be projected for the simulated samples. The number of samples for which pre-1951 earnings are needed is based on the age and earnings pattern of the worker as of 1988. For workers age 62 or older in 1988, pre-1951 earnings are projected for each corresponding simulated sample. For workers younger than age 62 in 1988, pre-1951 earnings are projected only for a fraction of the corresponding simulated samples-zero is assumed for all subsequent samples.

5. Other Considerations

The re-indexed widow(er) guarantee benefit calculation requires the date of birth of the aged or disabled widow(er). This information is not available in the original 1988 sample from the CWHS, but is found for survivor cases in the Master Beneficiary Record (MBR). Cross-referencing the CWHS with the MBR, and assuming a constant difference in age between the deceased worker and the surviving spouse provides this information for the simulated samples.

Starting in 1986, some benefit calculations may be affected by the Windfall Elimination Provision (WEP), which considers any noncovered pension of the worker. This provision was only partially effective by 1988, so only limited data existed in the CWHS and MBR. In the simulated samples, noncovered pension amounts may be assigned to some workers who show a pattern of steady earnings that is interrupted by an extended period of no earnings-which may be indicative of noncovered employment.

6. Benefit Calculations

Once the simulated samples are constructed, benefits can be calculated for each beneficiary in each sample. First, the model examines annual earnings and the corresponding quarters of coverage to determine if the insured status requirement is met for the particular benefit type.

If the insured status test is met, the benefit is computed for that record. Under the usual benefit calculation procedure, the Average Indexed Monthly Earnings (AIME) amount is found based on a specified number of highest years of indexed earnings-the number of "high-n" years of earnings depends on the year of eligibility of the worker. The eligibility year may precede the year of award, depending on either the year of attainment of age 62; the year of disability onset; or the year of death (survivor case). In the simulated samples, the relationship between the year of eligibility and the year of award is the same as the corresponding record in the 1988 sample.

Once the AIME for each record is computed, all relevant PIA formulas are applied and the highest applicable PIA becomes the PIA at award. Calculations that may apply include the following:

After the PIA is determined for each record in each simulated sample, the model computes the corresponding Monthly Benefit Amount (MBA). For retired workers, the MBA is either: (i) less than the PIA if retirement occurs before normal retirement age; (ii) equal to the PIA if retirement occurs at normal retirement age; or (iii) more than the PIA if retirement occurs after the normal retirement age. The reduction factors and delayed retirement credits will vary, depending on the worker's year of birth. Similar to the retired worker benefit, the MBA for an aged widow(er) is less than the PIA when benefit begins prior to the normal retirement age for the widow(er). For disabled workers, the MBA is assumed to be 100 percent of the PIA. For young survivor beneficiaries, the MBA is assumed to be 75 percent of the PIA.

Next, the average MBA for each type of beneficiary is calculated from the simulated samples for 2000-10. The average MBAs are converted to an index series, representing the year-over-year growth in benefits. Finally, the award amount for each beneficiary type is projected by applying the index to the actual average award for 2000.

7. Results

Table III.C1 indicates the award amount for female disabled workers will be roughly 27 percent less than male disabled workers throughout the next 10 years. Female benefits were as much as 33 percent less in the early 1980s. The gradual increase is primarily due to the increase in lifetime earnings of females, relative to those of males.

Table III.C2 shows award amounts for young and aged wives and husbands of disabled workers. The model assumes the amount of a spouse benefit is proportional to that of the worker. Future proportions are expected to be about the same as that for 2000. Note that the full benefit rate for a spouse is 50 percent of PIA, but the actual proportion over all beneficiaries is much less-roughly 17 percent for young spouses. This is mainly due to the Maximum Family Benefit (MFB) provision. The adjustment for the family maximum is made by proportionately reducing all auxiliary benefits until the total monthly benefits payable are within the maximum limit. Another example of reduced benefits to a spouse of a disabled worker is when the spouse is also entitled to a retired worker benefit. If the retirement benefit is smaller than the spouse benefit, only the difference is paid as a spouse benefit.

Table III.C3 shows award amounts for minor, disabled, and student children of disabled workers. The model assumes the amount of a child benefit is proportional to that of the worker. Future proportions are expected to be about the same as that for 2000. The full benefit rate for a child is 50 percent of PIA, but the MFB reduces the actual proportion as discussed above.

Table III.C4 shows award amounts for retired workers. The model projects retirement amounts based on the index of benefit growth from the simulated samples of awards. During the 1980s and early 1990s, award amounts to female retired workers decreased as a proportion of male award amounts. This is partly due to the lengthening computation period. Now that the computation period is at its maximum 35 years, women with longer and more complete earnings records result in more stable, albeit lower, benefits relative to men. Female retired worker benefits are projected to be roughly 65 percent of that for males throughout the short-range period.

Table III.C5 shows award amounts for survivors of deceased workers. The model projects these amounts based on the index of benefit growth from the simulated samples of awards for each of the following groups-11 types of beneficiaries in all:

Award amounts to aged widows are projected to remain the largest of any survivor award. Amounts for other types of survivors are lower because:

Award amounts to dependents of retired workers are not projected. See section E for a description of benefit projections for spouses and children of retired workers.

Contents Previous Next Glossary

List of Studies
December 26, 2001