Some guidelines need to be followed when you use PROC PANEL for analysis. For each cross section, PROC PANEL requires at least two time series observations with nonmissing values for all model variables. There should be at least two cross sections for each time point in the data. If these two conditions are not met, then an error message is printed in the log stating that there is only one cross section or time series observation and further computations will be terminated. You have to give adequate data for an estimation method to produce results, and you should check the log for any data related errors.
If the number of cross sections is greater than the number of time series observations per cross section, PROC PANEL while using PARKS method produces an error message stating that the phi matrix is singular. This is analogous to seemingly unrelated regression with fewer observations than equations in the model. To avoid the problem, reduce the number of cross sections.
Your data set could have multiple observations for each time ID within a particular cross section. However, PROC PANEL is applicable only in cases where you have only a single observation for each time ID within each cross section. In such a case, after you have sorted the data, an error warning specifying that the data has not been sorted in ascending sequence with respect to time series ID appears in the log.
The cause of the error is due to multiple observations for each time ID for a given cross section. PROC PANEL allows only one observation for each time ID within each cross section.
The following data set shown in Figure 20.2 illustrates the preceding instance with the correct representation.
Figure 20.2: Single Observation for Each Time Series
Obs | firm | year | production | cost |
---|---|---|---|---|
1 | 1 | 1955 | 5.36598 | 1.14867 |
2 | 1 | 1960 | 6.03787 | 1.45185 |
3 | 1 | 1965 | 6.37673 | 1.52257 |
4 | 1 | 1970 | 6.93245 | 1.76627 |
5 | 2 | 1955 | 6.54535 | 1.35041 |
6 | 2 | 1960 | 6.69827 | 1.71109 |
7 | 2 | 1965 | 7.40245 | 2.09519 |
8 | 2 | 1970 | 7.82644 | 2.39480 |
In this case, you can observe that there are no multiple observations with respect to a given time series ID within a cross section. This is the correct representation of a data set where PROC PANEL is applicable.
If for state ID 1 you have two observations for the year=1955, then PROC PANEL produces the following error message:
“The data set is not sorted in ascending sequence with respect to time series ID. The current time period has year=1955 and the previous time period has year=1955 in cross section firm=1.”
A data set similar to the previous example with multiple observations for the YEAR=1955 is shown in Figure 20.3; this data set results in an error message due to multiple observations while using PROC PANEL.
Figure 20.3: Multiple Observations for Each Time Series
Obs | firm | year | production | cost |
---|---|---|---|---|
1 | 1 | 1955 | 5.36598 | 1.14867 |
2 | 1 | 1955 | 6.37673 | 1.52257 |
3 | 1 | 1960 | 6.03787 | 1.45185 |
4 | 1 | 1970 | 6.93245 | 1.76627 |
5 | 2 | 1955 | 6.54535 | 1.35041 |
6 | 2 | 1960 | 6.69827 | 1.71109 |
7 | 2 | 1965 | 7.40245 | 2.09519 |
8 | 2 | 1970 | 7.82644 | 2.39480 |
In order to use PROC PANEL, you need to aggregate the data so that you have unique time ID values within each cross section. One possible way to do this is to run a PROC MEANS on the input data set and compute the mean of all the variables by FIRM and YEAR, and then use the output data set.