There Must Be a Better Way: Targets to Aim At.

In the previous post, Employment Metrics: There Must be A Better Way we noted some of the weaknesses with the press and blog coverage of the BLS Employment Report for Dec 2005.

Here are some suggested steps we might consider taking to improve the coverage. I recognize that these are easy to say and likely hard to do. Over the next few weeks, we will take a cut at following these principles ourselves (practicing what we preach) to see where they lead us.

Steps towards a better way of reporting on Employment trends

1. Multiple Metrics. Select a substantial number (at least 10) of different measures of the employment situation and report in one place on all of them at the same time.

The BLS collects and archives literally hundreds of different time series so there is plenty from which to chose the best ones. There are also other potential sources of time series (other than the BLS) that might supply some missing piece of the puzzle. Figuring out which measures are likely to be the most important and most revealing is a crucial first step towards our target of greater clarity.

Multiple, complementary measures provide vital checks and balances on the interpretation.

2. Use the Same Time Scale for each Trend Graph. Once you have the 10 or more of what we feel are the most important metrics, it's essential to use the same time scale to show the trend for each. It will also be helpful to those who will see your creation to understand why you selected that particular interval for review.

Even better, since no single time scale will give the complete picture, make each of the 10 plus time series charts available in at least two different time scale windows - e.g. 2000-2005 and 1990-2005.

3. Smooth the data to reveal the underlying patterns. Much of the BLS employment data is erratic. To figure out what's going on and present it in a visual fashion that others can understand normally involves invoking some sort of smoothing.

There is no simple formula for the analyst on how to do this. My suggestion and preference here is that rather than using the Seasonally Adjusted BLS numbers, that we start our analysis with the Unadjusted raw trend data for each month in the series.

From this is it then possible to calculate the Year over Year Change and Year over Year Percentage Change.

Using one of these 12 month change viewpoints will often provide some useful smoothing all by itself.

If the trends are still erratic and hard to read, then use some moving average period (e.g. 3 months or 6 months) to achieve further smoothing. Some trial and error may be required before you get an understandable picture.

Regarding smoothing, my experience tells me that you will know it when you see it.

4. Use at least 2 significant digits. Some data like the month to month percentage change in unemployment rate has for what seem to be historical reasons been presented with only a single significant digit. If something is changing by 0.0% or 0.1% in a month, important patterns may easily be overlooked.

5. Avoid monotonically increasing charts such as the picture of total number of non-farm workers. Compute the year over year change in these counts instead.

6. Share your data set. In order to carry out the steps above, you will have effectively created a new data set that never existed previously.

It will have your 10 or more metrics that you have decided are most important to understanding the situation.

You will have the values of these factors for 2 or more time intervals.

You may have carried out some calculations on your original source data to make it more willing to reveal its secrets (e.g. calculating year over year or moving average values), thereby creating additional factors.

And from this data set you will have created graphs that you feel best represent the underlying processes at work.

While your format for the data set may differ radically, what you will have is an object that is logically equivalent to a simple two-dimensional table where each row represents one time periods and each column represents one of the factors you have selected or calculated.

It's likely taken you a fair amount of work to create this data set. You have used it for your own analysis and reporting. This step is vital, because by sharing the same data with others, they may be able to discover patterns and insights that add to the ones you have already provided.

7. Make the Data Set "READILY REUSABLE". As best as you can, share you data in a format that makes it easy for the next person to begin where you left off with the minimum of time and effort. Time is almost everyone's most precious and scarce resource. When you do this step well, you enable the conversation and collaboration about the meaning of the data to continue.

One of the approaches that works brilliantly for this principle is to put your simple two-dimensional data set into a text file structured as a Comma Separated Value (CSV) file. CSV files satisfy the "readily reusable" requirement. Many existing personal computer applications (e.g. Microsoft Excel) can open such a file and make it ready for further analysis and charting.

In the examples that follow in the next few weeks, we will be aiming at applying as many of these principles as we can so you can see how it all works in practice.