2022 Projections Technical Documentation

Past projections

There is no technical documentation for clear reproducibility on the previous population and employment projections. It is difficult to say exactly which methods we used in the past, but it is a combination of using REMI and the Newling model (which focused on population density to predict growth). It is even less clear the past methodology when we consider that we produce different variations of the population/employment projections at a county, municipality, and then TAZ (Traffic analysis zone) level. It is my belief that we produced the county level projections at the county level through the use of REMI (either in-house or outsourced) and then utilized the “Newling model” completely separate from the REMI county level projections for the municipality/TAZ projections.

Based on what I understand, the last projections prior to 2022 were done by Bowen Liu and incorporated REMI (with land development simulation) at the county level and then used the Newling model to separately predict the municipality/TAZ level. To be explicit, if you summed the results of the Newling model produced municipality/TAZ level to the county level, they would not match the REMI level county projections. We also in the past did not try to “disaggregate” county level projections down to the municipality/TAZ level (which is a common method)

This is the most helpful summary of what was done (as stated before, there is no technical documentation for reproducibility for this method): (2019 Methodology)

Moving forward (2022)

Given the lack of reproducible documentation, we had to choose logical paths. The following is what was generally done as of July 2022. Each will be described in further detail in later sections.

• County Level Projections - REMI for county level population and employment projections
  • Did NOT include land use development simulation
• Municipality level projections - Disaggregation of REMI county level projections into municipalities using historical Census data for population and employment by municipality and using simple linear regression to predict the percent share for each municipality into the future.
• Travel Analysis Zone projections - TBD - we have not disaggregated the municipality level projections to the TAZ level. We may reach out to Penn DOT to have them disaggregate and use it in their travel demand model

From REMI, we worked with David Casazza (David.Casazza@remi.com) and Jeff Dykes (jeffrey.dykes@remi.com)

2022 County Level Projections Methodology

As stated above, county level projections for Lehigh and Northampton counties were taken out of REMI v4.5. We initially tried to replicate previous county level projections with land use development data (e.g. pipeline of the building of warehouses, apartments, etc.).

The previous 2019 REMI model simulation data is available: (2019 Simulation File). You can ask REMI to upload this XML file to your instance of REMI to see the prior simulation.

The reasons why we have not yet incorporated the land use development data into the county level projections via REMI:

• Drawbacks of simple, generalized modeling - The methodology described in this (2019 Methodology) to model the impact of building apartments, warehouses, and retail, etc. via construction costs, employee hiring, and sales are dependent on very broad assumptions (e.g. employees per square foot, sales per square foot, internet based statistics). If the modeling of the land use development projects are very generalized and not more realistic, they may over inflate the population and employment forecasts.

• Cost of time - If we are to improve the modeling of the land use development projects to make it more realistic, that is extremely time consuming as we may need to reach out to actual businesses and developers to understand more about their operations. We do not currently know whether that time investment will actually improve the accuracy of the forecasts.

• Accuracy of REMI forecasts without land development data - Baseline forecasts in REMI are out-of-the-box and require no additional preparation. They are the forecasts without any of the land use development simulation data. During the disaggregation of county level projections into municipality level forecasts (a process described in the next section), we tested the accuracy of the municipality level forecasts against Census municipality level population data and they were satisfactorily accurate. It is difficult to justify the additional work to include and model land use development projects, which may or may not improve the accuracy of forecasts. The process may be more reasonable with more than a single staff member.

2022 Municipal Level Projections Methodology

In contrast to previous projections, we utilized a disaggregation method of the REMI county level forecasts into each municipality. The general method was endorsed by the University of Tennessee Boyd Center and also used by Bob Diogo, NJTPA (New Jersey Transportation Planning Authority) in their projections.

• The University of Tennessee Boyd Center population projections methodology is located: (UT Methodology)

• Conversation with Bob Diogo confirming our methodology: (Bob Diogo NJTPA)

We applied the same methodology for the disaggregation of population and employment, these Excel based models are found here:

• 2022 Municipality Population Model (Population Disaggregation)

• 2022 Municipality Employment Model (Employment Disaggregation)

Navigation around the Excel Disaggregation Model

Each grouping of tabs in both the population and employment disaggregation excel models are organized by “yellow” tabs either “Summary”, “Model”, “Raw Data”, etc.

The most important groupings of tabs are the “Model” and “Raw Data” tabs.

Each tab within the yellow “Model” tab has a tab for municipal forecasts for “Lehigh County” and also “Northampton County” towns/cities.

Both the “Lehigh County” and “Northampton County” forecast tabs have 5 steps to disaggregate the county level REMI forecasts. These steps are clearly listed in these sheets.

Model Tab (applies to both population and employment models)

Step 1: Retrieve Subcounty Historical Estimates (for either population or employment)

For the population and employment disaggregation models for example, we need the historical population and employment estimates by municipality in order to infer about the future growth. See Excel formula.

For population, see Census Subcounty Resident Population Estimates data. For employment, refer to Census OnTheMap Employment by City/Town.

Within Step 1, the historical data was inputted for every city/town in that specific county. As of 2022, the historical data was 2010-2020, so populate accordingly.

Step 2: Calculate Subcounty Proportions of Total (what is the percent of total of each muncipality)

Calculate the % of total of each city/town for each year. See Excel formula.

Step 3: Create Linear Trendline of Subcounty Share of Total (Predicting percentage of total for each municipality going into the future using linear regression)

Based on the historical % of total for each municipality, draw the best fit line through the % of total estimates for each municipality, in order to see how it might change in the next 10-20 years. See Excel formula.

Step 4: Retrieve County-Level Projections (These are the REMI county level forecasts)

Bring in the county level forecasts generated from REMI into this section. We will apply the municipal level historical percent share (Step 1) and forecasted percent share (Step 3) to these county level forecasts to get the disaggregation in the next step. See Excel formula.

Step 5: Calculate Subcounty Numbers as a Proportion of County Projected Numbers Total (disaggregate county forecasts into municipality forecasts)

For the historical years (2010-2020), multiply Step 1 and Step 4 for each municipality. For forecasted years (2021 - 2050), multiply Step 3 and Step 4. Now you have the disaggregated municipality projections.

Summary Tabs

The tabs in this section are summarizing the disaggregated municipality numbers calculated in the Model tabs. There may be some graphs and summary tabs.

Testing the accuracy of forecasts

The accuracy of the municipality level forecasts were only tested for the population model. The reason for this is the availability and compatibility of the historical Census population data by city/town.

We were able test the accuracy of the population disaggregation model by “pushing back” the historical years and trying to see how our model predicts the population by city/town (municipality) compared to the Census city/town estimates.

For example, we know that the historical data currently is 2010-2020, but if we simply say that historical data for testing purposes is now 2005-2014, we can see how the model predicts population growth by municipality compared to Census estimates for the years of 2015-2020.

In the population model, the tabs to test the accuracy are named “YYYY Comparison” (e.g. 2015 Comparison). In column G, we show how far we were off from the Census estimates. In column I, we compared how accurate this model was to the previous Newling model (done previous to 2022).

Given that Census has a margin of error of 10% and predicting changes in smaller areas (cities/towns) are more difficult than the county level, the accuracy tests were acceptable.

It should be mentioned that 2020 was an outlier year and would have been difficult to predict something like a pandemic causing changes in population in municipalities. Therefore, 2020 accuracy should be taken with a grain of salt.

Accuracy of employment municipality forecasts

We did not measure the accuracy of the disaggregated employment municipality forecasts because of the mismatch in historical city/town employment data from Census OnTheMap and the Bureau of Economic Analysis employment data by county, which REMI uses.

We apply the same methodology to disaggregate the employment county level forecasts as the population county level forecasts, which is taking historical data by city/town (municipality), calculating the % of total, and multiplying it against the county level forecasts to get municipal level forecasts.

The issue is that Census OnTheMap and Bureau of Economic Analysis (BEA) calculate “employment” in different ways, causing issues if we were to measure “accuracy”. If you were to summarize the Census OnTheMap employment by city/town into the county level and compare the employment county numbers against what the BEA/REMI states, there would be a large disparity.

See this message from Kim DeLellis from Center for Workforce Information and Analysis (CWIA) about the difference in calculation of employment numbers between Census and BEA: (Question on Employment Calculation Disparity from CWIA)

BEA does not calculate municipality level employment, only the Census does. Bob Diogo from NJTPA also uses the same employment by municipality data from the Census for his projections. There is nothing else available.

Writing the actual report

The report is made in R Markdown to produce a pdf. It is located here: (2022 Report)

It would make sense to replicate each future report using this as a template as all the graphs and tables just need new data. There is a folder named “Raw Data” in the (2022 Report Folder). You may need to confirm/change the working directories to make sure it is pulling the data from the “Raw Data” folder.