Hi Todd, there are a couple different ways to conceptualize a population synthesizer using various Census datasets.  For details, I would refer you to a couple of papers I co-authored:

 

Rousseau, Guy and John Bowman. Validation of the Atlanta (ARC) Population Synthesizer (PopSyn).  Paper prepared for the TRB Conference on Innovations in Travel Modeling, Austin, Texas, May 21- 23, 2006, http://www.trb.org/conferences/tdm/papers/BS1B%20-%20Bowman%20and%20Rousseau%20ARC%20PopSyn.pdf

 

Rousseau, Guy and Greg Erhardt. Use of Census PUMS Data in Activity Based Models. Paper prepared for the USDOT, CTPP Status Report Newsletter, May 2008 http://www.fhwa.dot.gov/ctpp/sr0508.htm

 

Our ARC population synthesis procedure takes into account zonal and regional controls and includes a procedure to allocate households to sub-zones.  The ARC population synthesizer was developed to be a flexible tool for creating synthetic populations for Activity-Based modeling.  The population synthesizer takes as an input Census data and zonal-level and regional marginal distributions of households by various characteristics that are used as controls which the synthetic population is forced to match.  The population synthesizer first develops a “base year” population distribution using year 2000 Census data.  A set of controlled for attributes are defined, and Census Summary File 1, Summary File 3, and the Census Transportation Planning Package information is used to develop single and multi-dimensional distributions of these attributes.  These attributes include:

 

·         Householder age

·         Household size

·         Household income

·         Presence of children in household

·         Number of workers in household

·         Number of units in household structure

·         Population race

·         Population group quarters type

 

Once this distribution is established, the population synthesis tool then samples PUMS records to create a fully enumerated representation of the population.  In order to use the ARC model to forecast travel, it is necessary to develop future year synthetic populations.  The population synthesizer updates the base year distributions of household and person attributes to provide future year distributions of these attributes, based on future year data found in ARC’s regional model TAZ inputs.  The population synthesizer is applied at the TAZ level for the entire model area.  Below is a schematic of the overall process, all programmed in JAVA:

 

 

I hope this helps.  Let me know if you have any questions.

 

Thanks,

 

Guy

 

Guy Rousseau

Modeling Manager

Atlanta Regional Commission

 

 

From: ctpp-news-bounces@chrispy.net [mailto:ctpp-news-bounces@chrispy.net] On Behalf Of Graham, Todd
Sent: Friday, March 27, 2009 3:29 PM
To: 'ctpp-news@chrispy.net'
Subject: [CTPP] Small-area socioeconomic data sources, synthesis techniques

 

A few questions for anyone who'd care to offer opinions...

 

Who on the list has experience "synthesizing" small-area (= TAZ Zones or Tracts or Block Groups) socioeconomic crosstabs that go beyond the tables published by Census Bureau?  By "beyond," I'm specifically thinking of using multiple households characteristics as segmenting dimensions.

 

And how do you go about it?  Iterative proportional fitting – or propensity? or some other joint distribution technique??

 

About my own objectives: 

For analyses and forecasts-prep at Metropolitan Council, we would like to segment our region's Census 2000 households on multiple dimensions:

 age of householder (4 categories) X household size (4 categories) X household income (5 categories) X geographic units (1200 Zones).

 

I’ve been thinking that we would start with CTPP table 64 (crosstab of household size X household income X TAZ Zones) and then synthesize the additional segmentation by age group (drawing on PUMS data).  But I’m skeptical that household size X household income is enough information to predict age of householder.  Any thoughts??

 

Alternately… Census SF3 table P55 offers crosstab of age group X household income X Block Groups.  Perhaps start there and synthesize the additional segmentation by household size.  This might work better! – but then there’s the additional hassle of correspondence between Block Groups vs TAZ Zones. 

 

Any advice is appreciated.  Thanks.

 

 

-- Todd Graham

   Metropolitan Council Research

   651/602-1322