Observations (Temporal Causal Modeling)

On the Fields tab, use the Observations settings to specify the fields that define the observations.

Note: If the active dataset has a date specification, then the observations are defined by the date specification and cannot be modified in the temporal causal modeling procedure. Date specifications are created from the Define Dates dialog or the DATE command.
Observations that are defined by date/times

You can specify that the observations are defined by a field with a date, time, or datetime format, or by a string field that represents a date/time. String fields can represent a date in YYYY-MM-DD format, a time in HH:MM:SS format, or a datetime in YYYY-MM-DD HH:MM:SS format. Leading zeros can be omitted, in the string representation. For example, the string 2014-9-01 is equivalent to 2014-09-01.

In addition to the field that defines the observations, select the appropriate time interval that describes the observations. Depending on the specified time interval, you can also specify other settings, such as the interval between observations (increment) or the number of days per week. The following considerations apply to the time interval:
  • Use the value Irregular when the observations are irregularly spaced in time, such as the time at which a sales order is processed. When Irregular is selected, you must specify the time interval that is used for the analysis, from the Time Interval settings on the Data Specifications tab.
  • When the observations represent a date and time and the time interval is hours, minutes, or seconds, then use Hours per day, Minutes per day, or Seconds per day. When the observations represent a time (duration) without reference to a date and the time interval is hours, minutes, or seconds, then use Hours (non-periodic), Minutes (non-periodic), or Seconds (non-periodic).
  • Based on the selected time interval, the procedure can detect missing observations. Detecting missing observations is necessary since the procedure assumes that all observations are equally spaced in time and that there are no missing observations. For example, if the time interval is Days and the date 2014-10-27 is followed by 2014-10-29, then there is a missing observation for 2014-10-28. Values are imputed for any missing observations. Settings for handling missing values can be specified from the Data Specifications tab.
  • The specified time interval allows the procedure to detect multiple observations in the same time interval that need to be aggregated together and to align observations on an interval boundary, such as the first of the month, to ensure that the observations are equally spaced. For example, if the time interval is Months, then multiple dates in the same month are aggregated together. This type of aggregation is referred to as grouping. By default, observations are summed when grouped. You can specify a different method for grouping, such as the mean of the observations, from the Aggregation and Distribution settings on the Data Specifications tab.
  • For some time intervals, the additional settings can define breaks in the normal equally spaced intervals. For example, if the time interval is Days, but only weekdays are valid, you can specify that there are five days in a week, and the week begins on Monday.
Observations that are defined by periods or cyclic periods

Observations can be defined by one or more integer fields that represent periods or repeating cycles of periods, up to an arbitrary number of cycle levels. With this structure, you can describe series of observations that don't fit one of the standard time intervals. For example, a fiscal year with only 10 months can be described with a cycle field that represents years and a period field that represents months, where the length of one cycle is 10.

Fields that specify cyclic periods define a hierarchy of periodic levels, where the lowest level is defined by the Period field. The next highest level is specified by a cycle field whose level is 1, followed by a cycle field whose level is 2, and so on. Field values for each level, except the highest, must be periodic with respect to the next highest level. Values for the highest level cannot be periodic. For example, in the case of the 10-month fiscal year, months are periodic within years and years are not periodic.

  • The length of a cycle at a particular level is the periodicity of the next lowest level. For the fiscal year example, there is only one cycle level and the cycle length is 10 since the next lowest level represents months and there are 10 months in the specified fiscal year.
  • Specify the starting value for any periodic field that does not start from 1. This setting is necessary for detecting missing values. For example, if a periodic field starts from 2 but the starting value is specified as 1, then the procedure assumes that there is a missing value for the first period in each cycle of that field.
Observations that are defined by record order
For column-based data, you can specify that the observations are defined by record order so that the first record represents the first observation, the second record represents the second observation, and so on. It is then assumed that the records represent observations that are equally spaced in time.