Description: <DIV STYLE="text-align:Left;"><DIV><DIV><P STYLE="margin:0 0 11 0;"><SPAN STYLE="font-style:italic;"><SPAN>Data Source: </SPAN></SPAN><SPAN><SPAN>The primary data source used for this analysis are point-level business establishment data from InfoUSA. This commercial database produced by </SPAN></SPAN><SPAN STYLE="font-style:italic;"><SPAN>InfoGroup </SPAN></SPAN><SPAN><SPAN>provides a comprehensive list of businesses in the SCAG region, including their industrial classification, number of employees, and several additional fields. Data have been post-processed for accuracy by SCAG staff and have an effective date of 2016. </SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN STYLE="font-style:italic;"><SPAN>Locally-weighted regression: </SPAN></SPAN><SPAN><SPAN>First, the SCAG region is overlaid with a grid, or fishnet, of 1km, 2km, and ½-km per cell. At the 1km cell size, there are 16,959 cells covering the SCAG region. Using the Spatial Join feature in ArcGIS, a sum total of business establishments and total employees (i.e., not separated by industrial classification) were joined to each grid cell. Note that since cells are of a standard size, the employment total in a cell is the equivalent of the employment density. A locally-weighted regression (LWR) procedure was developed using the R Statistical Software package in order to identify subcenters</SPAN></SPAN><SPAN STYLE="font-style:italic;"><SPAN>.</SPAN></SPAN><SPAN><SPAN>The below procedure is described for 1km grid cells, but was repeated for 2km and 1/2km cells. </SPAN></SPAN></P><OL STYLE="margin:0 0 0 0;padding:0 0 0 0;"><LI><P STYLE="margin:0 0 0 0;"><SPAN><SPAN>Identify local maxima candidates.</SPAN></SPAN><SPAN><SPAN>Using R’s lwr package, each cell’s 120 nearest neighbors, corresponding to roughly 5.5 km in each direction, was explored to identify high outliers or </SPAN></SPAN><SPAN STYLE="font-style:italic;"><SPAN>local maxima </SPAN></SPAN><SPAN><SPAN>based on the total employment field. Cells with a z-score of above 2.58 were considered local maxima candidates.</SPAN></SPAN></P></LI><LI><P STYLE="margin:0 0 0 0;"><SPAN><SPAN>Identify local maxima</SPAN></SPAN><SPAN><SPAN>. LWR can result in local maxima existing within close proximity. This step used a .dbf-format spatial weights matrix (knn=120 nearest neighbors) to identify only cells which are higher than all of their 120 nearest neighbors. At the 1km scale, 84 local maxima were found, which will form the “peak” of each individual subcenter. </SPAN></SPAN></P></LI><LI><P STYLE="margin:0 0 0 0;"><SPAN><SPAN>Search adjacent cells to include as part of each subcenter</SPAN></SPAN><SPAN><SPAN>. In order to find which cells also are part of each local maximum’s subcenter, we use a queen (adjacency) contiguity matrix to search adjacent cells up to 120 nearest neighbors, adding cells if they are also greater than the average density in their neighborhood. A total of 695 cells comprise subcenters at the 1km scale. </SPAN></SPAN></P></LI></OL><P STYLE="margin:0 0 11 0;"><SPAN /></P><P STYLE="margin:0 0 11 0;"><SPAN>A video from Kane et al. (2018) demonstrates the above aspects of the methodology (please refer to 0:35 through 2:35 of </SPAN><A href="https://youtu.be/ylTWnvCCO54"><SPAN><SPAN>https://youtu.be/ylTWnvCCO54</SPAN></SPAN></A><SPAN>), with several minor differences which result in a different final map of subcenters: different years and slightly different post-processing steps for </SPAN><SPAN STYLE="font-style:italic;"><SPAN>InfoUSA</SPAN></SPAN><SPAN>data, video study covers 5-county region (Imperial county not included), and limited to 1km scale subcenters.</SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>A challenge arises in that using 1km grid cells may fail to identify the correct local maximum for a particularly large employment center whose experience of high density occurs over a larger area. The process was repeated at a 2km scale, resulting in 54 “coarse scaled” subcenters. Similarly, some centers may exist with a particularly tightly-packed area of dense employment which is not detectable at the medium, 1km scale. The process was repeated again with ½-km grid cells, resulting in 95 “fine scaled” subcenters. In many instances, boundaries of fine, medium, and coarse scaled subcenters were similar, but differences existed. </SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>The next step was to qualitatively comparing results at each scale to create the final map of 72 job centers across the region. Most centers are medium scale, but some known areas of especially employment density were better captured at the 2km scale while . Giuliano and Small’s (1991) “ten jobs per acre” threshold was used as a rough guide to test for reasonableness when choosing a larger or smaller scale. For example, in some instances, a 1km scale included much additional land which reduced job density well below 10 jobs per acre. In this instance, an overlapping or nearby 1/2km scaled center provided a better reflection of the local employment peak. Ultimately, the goal was to identify areas where job density is distinct from nearby areas. </SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>Finally, in order to serve land use and travel demand modeling purposes for Connect SoCal, job centers were joined to their nearest TAZ boundaries. While the identification mechanism described above uses a combination of point and grid cell boundaries, the job centers boundaries expressed in this layer, and used for Connect SoCal purposes, are built from TAZ geographies. In Connect SoCal, job centers are associated with one of three strategies: focused growth, coworking space, or parking/AVR.</SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN STYLE="font-weight:bold;">Data Field/Value description:</SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>name: Name of job center based on name of local jurisdiction(s) or other discernable feature.</SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>Focused_Gr: Indicates whether job center was used for the 2020 RTP/SCS Focused Growth strategy, 1: center was used, 0: center was not used.</SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>Cowork: Indicates whether job center was used for the 2020 RTP/SCS Co-working space strategy, 1: center was used, 0: center was not used.</SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>Park_AVR: Indicates whether job center was used for the 2020 RTP/SCS parking and average vehicle ridership (AVR) strategies, 1: center was used, 0: center was not used. </SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>nTAZ: number of Transportation Analysis Zones (TAZs) which comprise this center.</SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>emp16: Estimated number of workers within job center boundaries based on 2016 InfoUSA point-based business establishment data. Values are rounded to the nearest 1000. </SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN><SPAN>acres: Land area within job center boundaries based on grid-based identification mechanism (i.e., not based on TAZ boundaries shown). Values are rounded to the nearest 100. </SPAN></SPAN></P><P STYLE="margin:0 0 11 0;"><SPAN /><SPAN /></P><P STYLE="margin:0 0 11 0;"><SPAN /><SPAN /></P></DIV></DIV></DIV>
Service Item Id: 06c4f5f744f54354b05707be5246778b
Copyright Text: SCAG
Additional references: Giuliano, Genevieve, and K. A. Small. 1991. Subcenters in the Los Angeles region. Regional Science and Urban Economics 21, p. 163-182.
Kane, K., Hipp, J. R., & Kim, J. H. 2018. Los Angeles employment centers in the twenty-first century. Urban Studies 55:4, 844-869. See https://cloudfront.escholarship.org/dist/prd/content/qt3d55p4zt/qt3d55p4zt.pdf