SOI Tax Stats - Partnership Study Data Sources and Limitations

 

This page contains information about data sources and limitations for SOI's annual Partnership Study.


Tax Year 2006 statistics are estimates based on a stratified probability sample of 45,346 returns selected from a population of 3,164,232 partnerships.  Tax Year 2006 covers returns processed by the IRS during Calendar Year 2007.  All partnerships engaged in business in, or having income from sources within, the United States were required to file either a Form 1065, U.S. Partnership Return of Income, or a Form 1065-B, U.S. Return of income for Electing Large Partnerships, to report income or loss, deductions, tax credits, and other tax-related items generated by the partnership. The statistics are only for active partnerships, which are defined as those that reported any items of income or deduction derived from a trade or business, or from rental or portfolio income.

The population was stratified into classes based on industry, type of return, size of total assets, and size of certain receipt or income amounts from both ordinary business income (loss) and portfolio income (loss). Returns were selected from these classes at various probabilities ranging from 0.07 percent to 100 percent, and were weighted to represent the total population. There were 1,346 sampled returns that were ruled “out of scope” because they had no activity or duplicated other returns already in the study. This resulted in a final sample of 44,000 returns, and an estimated overall active population of 2,947,116.

Because the data presented in this article are based on a sample of returns, they are subject to sampling error. To properly use the data, the magnitude of the potential sampling error needs to be known. Coefficients of variation (CVs), the ratio of an estimate's standard error to the estimate, are used to measure this magnitude. Figure KPDF presents the coefficients of variation for certain money amounts, for selected industrial sectors.  The estimate is judged more reliable when the coefficient of variation is smaller.