We have now received three reviews for your paper "Agglomeration economies, accessibility and the spatial choice behavior of relocating firms" submitted to the Journal of Transport and Land Use. All three reviews (attached below) concluded "revisions required", and made many suggestions about the revisions that would be necessary to secure acceptance. Based on these reviews, I would encourage you to revise and resubmit the paper. Some additional comments I have on reading the paper (1) feel free to use color in the maps, as this is published online; (2) try to make sure you clarify the use of the term "accessibility", in some cases it seems to mean urbanization economies, in others it is proximity to infrastructure. (1) B/W MAPS ARE REPLACED WITH COLOR MAPS (2) THE USE OF ACCESSIBILITY AND URBANISATION ECONOMIES IS REVISED, AND THE DISTINCTION BETWEEN URBANIZATION ECONOMIES AND PROXIMITY TO INFRASTRUCTURE SHOULD BE CLEAR NOW Thanks you for your contribution to JTLU and we look forward to seeing a revised paper in the near future. Sincerely, David Levinson ------------------------------------------------------ Reviewer A: This is an innovative and interesting paper. In general this paper is based on a sound analysis and is well written. However, some issues need to be addressed After revisions this paper can be accepted for publication Agglomeration economies, accessibility and the spatial choice behaviour or relocating firms The aim paper of this paper is to analyze at the firm level the impact of accessibility, agglomeration economies and firm characteristics on relocation decisions of firms. This is an innovative and interesting subject. In general this paper is based on a sound analysis and is well written. However, some issues need to be addressed: • The introduction is rather long and shows some overlap between the paragraphs: o P.2, final paragraph: the focus of the paper is mentioned. o P.3, first paragraph: in different wordings the same focus is mentioned o P.3, final paragraph: the focus is mentioned again I would like to advice the author to rewrite and shorten the introduction. - THE INTRODUCTION HAS BEEN REWRITTEN AND OVERLAP HAS BEEN REMOVED • Context accessibility is not a common used concept. In itself accessibility is already contextual. - INDEED. THE USE OF ‘CONTEXT ACCESSIBILITY’ IS AVOIDED • I was wondering whether congestion levels were taken into account. These can have a large impact on accessibility in rush hours. - YES. THIS COMMENT IS NOW INCLUDED IN A MORE DETAILED DESCRIPTION IF DATA THAT WAS USED • The conclusions are rather short. I was wondering whether the author could reflect on their analysis. I also would like to see some suggestions for future research. For example, one of the conclusions is that accessibility of limited influence is. That result might be related to the chosen accessibility indicators (proximity to transport access points and log sum accessibility). The proximity measure is a very simple one and is sensitive for distances. The log sum for business trips and especially commuting trips hardly take into account constraints of activity and travel patterns. - THE CONCLUSIONS IS EXTENDED. THE USE OF THE PROXIMITY MEASURE IS REFLECTED UPON AND SUGGESTION FOR IMPROVEMENTS AND FUTURE RESEARCH ARE MADE. ------------------------------------------------------ ------------------------------------------------------ Reviewer B: “Agglomeration economies, accessibility and the spatial choice behavior of relocating firms” The authors apply a two-step model to firm micro level data for the Dutch province of South-Holland to determine the extent to which relocation decisions are dependent on accessibility and agglomeration economies, after controlling for firm characteristics. The authors first estimate the relocation probability of firms using a binary regression model, and then test for the importance of individual firm attributes, local agglomeration economies and accessibility attributes of the preferred locations on the observed location decisions. The results show that firm relocation behavior is more affected by firm level attributes than by agglomeration economies and accessibility to transport networks. Section 1: Introduction Comment 1: It is not clear to which empirical literature the authors are referring to in page 1. The empirical literature on agglomeration economies is older than the 1990s, do the authors have in mind the empirical literature on firm location? It needs to be made precise in the sentence. - THE TEXT IS REVISED TO MAKE IT MORE PRECISE Comment 2: There is a general confusion in relation to the methodologies found in the literature estimating agglomeration economies. This is problematic for some of the comments made by the authors. In page 2, the authors claim that there is lack of firm-level evidence in the agglomeration literature, but this not true anymore for studies estimating the effect of agglomeration externalities on productivity, whereas it appears to still be true for the studies interested in firm location. - TEXT IS REVISED TO MAKE THIS MORE CLEAR (FOOTNOTE 1 IS ADDED TO NOTE THIS ISSUE) Comment 3: What are accessibility economies? (in Abstract and Introduction, page 4)- THIS WAS NOT A PROPER TERM AND IS AVOIDED. THE REVISED ARTICLE NOW REFERS TO AGGLOMERATION AND ACCESSIBILITY EXTERNALITIES Section 2: conceptual framework Comment 1: In general, this section is somewhat haphazard and should be improved by being reorganized and reduced. - STRUCTURE OF SECTION IS REVISED Comment 2: Most of the section is focused on the empirical literature on agglomeration economies, in contrast to the near absence of discussion of the findings from the industrial organization literature. This is even more surprising in light of the results obtained: the main factors influencing firm relocation behaviour are related to the characteristics of firms, rather than the attributes of the context to where firms relocate. - THE REVISED TEXT SHOULD PUT THE ANALYSIS IN THE PROPER PERSPECTIVE (SEE NEW SUBSECTION ‘FROM FIRM LEVEL ATTRIBUTES TO EXTERNALITIES’) Comment 3: Since the authors organize their empirical analysis around 5 hypotheses, it would be interesting to know what previous research has concluded in relation to each of the stated hypothesis. - THE HYPOTHESIS IS NOW INTRODUCED IN THE APPROPRIATE PARAGRAPHS. THIS GIVES THE SECTION A MORE LOGICAL LINE OF REASONING. INHERENTLY, IT IS MORE CLEAR NOW WHAT THE LITERATURE HAS TO SAY ABOUT THESE HYPOTHESIS. Section 3: modelling framework Comment 1: The discussion of the construction of the transport and accessibility variables is unclear in relation to the data used. - THIS DESCRIPTION HAS BEEN REWRITTEN TO MAKE IT MORE PRECISE AND EFFECTIVE Comment 2: It is unclear why from the 1 million observations available, the authors say they use only 100,000. It is also unclear which criteria were used to select only five economic sectors when the raw data available presumably covers the whole economy. - THE ESTIMATIONS WERE MADE FOR A SELECTION OF INDUSTRIES. NOW THE PAPER INCLUDES ON OVERVIEW OF THE NUMBER OF OBSERVATIONS THAT WAS USED IN EACH ESTIMATION DATASET Comment 3: How exactly was the “back casting” done to obtain past measures of accessibility? - THE BACKCASTING STUDY WAS EXECUTED BY THE OWNER OF THE NATIONAL TRANSPORT MODEL (THE TRANSPORT RESEARCH CENTRE FROM THE MINISTRY OF TRANSPORT AND PUBLIC WORKS). THE BACKCASTING STUDY WAS MADE ON HISTORIC NETWORKS, COUNTS, AND SOCIO ECONOMIC DATA FOR THE BACKCASTIGN YEARS. OUR ANALYSIS ONLY USES THE OUTCOMES. A DESCRIPTION OF THAT ANALYSIS IS BEYOND THE SCOPE OF THIS PAPER. Comment 4: Some conclusions are drawn and used as reasons as to why some controls need not be included in the models (e.g., wages, page 16). Can the authors provide some values supporting the statements they make? In particular for the statement that wages vary little within sectors across the Province: “(…) wage rates within a sector are uniform and there is no need to control for labour force characteristics such as level of education, …”. - GIVEN THE SMALL SIZE OF THE STUDY AREA (ONE OF THE PROVINCES IN THE WEST OF THE NETHERLANDS - FUNCTIONALLY PART OF ONE LARGER URBAN REGIO- AND ALL LOCATIONS WITHIN THE STUDY AREA ARE WITHIN A REASONABLE COMMUTING DISTANCE) THE ABSENCE OF DIFFERENCES IN WAGE RATES WITHIN SECTORS IS OBVIOUS. TO US IT DOESN’T SEEM NECESSARY TO PROVIDE ADDITIONAL FIGURES. Comment 5: Page 14. It is not clear whether the data available is yearly from 1988 to 1997 or just for the two extremes. In table 2 there are 249,717 establishments: how many in each year? If it is a panel data context how did the estimator used account for with-firm (or in this case within-establishment) variation, given that a firm can have more than one establishment. - THE NEW TABLE (TABLE 1) SHOULD SOLVE THIS UNCLARITY Section 4: results Comment 1: What is the quantitative interpretation of the coefficients in tables 3-7?- THE MEANING OF THE COEFFICIENTS IS MENTIONED, AND THE RESULTS ARE DISCUSSED IN MORE DETAIL Comment 2: Hypothesis 3, page 20-21. The discussion of the findings for the influence from infrastructure proximity is misleading. Looking at the tables, the only sector that has a significant coefficient is manufacturing and only for the beta location, near to train stations and highway onramp. According to your results, there is no significant effect from any type of infrastructure proximity for the remaining sectors. This also applies to the respective conclusion in page 24, section 5. - THIS CONCLUSIONS REGARDING THIS HYPOTHESIS IS REWRITTEN: CONCLUSIONS ARE MADE WITH MORE PRUDENCE, AND THE DISCUSSION IS MORE ELABORATE Comment 3: There is no discussion of the measures of goodness of fit (not even a definition of the measures shown in the tables) of the models estimated in any of the analyses. No discussion of issues of heteroskedasticity or multicollinearity. Are the standard errors robust to heteroskedasticity or any type of intra-group correlation? - MULTICOLLINEARITY BETWEEN ATTRIBUTES IN THE UTILITY FUNCTIONS WAS AVOIDED BY TESTING THE CORRELATION BETWEEN LOCATION ATTRIBUTES. THERE WAS NO SIGNIFICANT CORRELATION BETWEEN THE LOCATION ATTRIBUTES IN THE MODELS. SOME EXAMPLES: THE DUMMY ATTRIBUTES FOR PROXIMITY TO HIGHWAY AND/OR RAILWAY STATIONS SOLVED CORRELATION BETWEEN THE ABSOLUTE DISTANCE TO HIGHWAY ONRAMP AND TRAIN STATION (THE HIGHWAY AND RAIL NETWERK IS BUNDLED). THE LOGSUMS FOR COMMUTING AND BUSINESS TRIPS WERE HIGHLY CORRELATED. TO TEST FOR URBANISATION ECONOMIES WE COULD ONLY USE THE SUM OF THESE TWO MEASURES (OR INCLUDE ONLY ONE OF THE TWO, BUT THE SUM WAS PREFERRED). HETEROSKEDASTICITY OF STANDARD ERRORS CAN BE THE RESULTS OF UNOBSERVED VARIATIONS IN PREFERENCES OF THE DECISION MAKERS OR INTRAGROUP CORRELATION OF THE CHOICE ALTERNATIVES. THE MODELS THAT WERE APPLIED DEAL WITH THIS AS FAR AS POSSIBLE. IN SPATIAL CHOICES IT IS DIFFICULT TO DISTINGUISH GEOGRAPHIC CLUSTERS IN A CONTINUOUS URBAN ENVIRONMENT. THE COMPETING DESTINTION MODEL INCLUDES A PROXY MEASURE FOR SPATIAL GROUP MEMEBERSHIP THROUGH THE CENTRAILTY PARAMETER. THIS PARAMETER WAS SIGNIFICANT, SHOWING THE NEED TO ACCOUNT FOR SPATIAL CLUSTERING. THE VARIATION IN LOCATION PREFERENCES OF YOUNG OR GROWING FIRMS WAS TESTED WITH INTERACTION TERMS BUT THESE INETERATION TERMS DID NOT LEAD TO SIGNIFICANT BETTER MODELS. Other: 1. Section 2 has very long paragraphs that make the reading difficult. Also the passages between paragraphs do not link well. Suggest re-organizing and reducing this section. – THIS SECTION IS RESTRUCTURED 2. Size of font in tables and sometimes in the text is not the same. There are typos in the name of industries in tables. - MANUFACTERING IS REPLACED WITH MANUFACTURING. TETA IS REPLACED WITH THETA. FONT TYPE IS CHANGED TO FONT USED IN ARTICLE. FONT SIZE COULD NOT BE CHANGED TO KEEP THE TABLES FIT INTO THE PAGE 3. The superscripts in the equations are not defined in the text. -IS NOW INCLUDED IN TEXT 4. References in the reference list are not consistent in terms of style. - HAS BEEN REVISED 5. Authors should discuss in more detail the policy and planning implications of their results. - THE CONCLUSIONS IS EXTENDED WITH A DISCUSSION OF HOW THE RESULTS CAN BE USED TO SUPPORT URBAN PLANNING ------------------------------------------------------ ------------------------------------------------------ Reviewer C: ------------------------------------------------------ Review of: #125 Agglomeration economies, accessibility and the spatial choice behavior of relocating firms I would rate this paper Very Good, and recommend it be published subject to the revisions I outline below. First, the model is straightforward, but I would recommend that a full version of the equation be entered following equation #4 which has the components of the equations V and C spelled out. - V AND C ARE SPELLED OUT IN EQUATIONS 2 AND 3 It is minor, but the authors employ European style numbering in the tables (i.e. commas for decimals). This must be ruthlessly opposed at every turn ☺ - US STYLE NUMBERING IS RUTHLESSLY APPLIED IN ALL TABLES ;-) I would like to see more explanation of the results, beyond simple interpretation of the coefficients. On page 18 for example these very important results should be explained not only in their statistical significance, but also the magnitude of the effects. These are probabilities and are a fairly important piece of the puzzle, as very small impacts can be statistically meaningful, but have no real policy relevance. The authors have not made this explanation in the paper. - THE DISCUSSION OF RESULTS HAS BEEN WRITTEN WITH MORE DETAIL I am also concerned that the comparison of these findings to other studies is weak, and needs to go beyond simply citing the other papers with similar findings. I would expect to see comparisons and magnitudes from other studies compared to the findings in this literature. This is an stark omission in several areas. - A COMPARISON OF MAGNITUDES FROM OTHER STUDIES IMPLIES ALMOST A META-ANALYSIS WHICH IS BEYOND THE SCOPE OF THIS PAPER On page 19, where the authors discuss the estimated parameters, I am not clear what the parameter estimates represent (probabilities or simply the estimated beta which are different things). - THE MEANING OF COEFFICIENTS IN THE RESULTS SECTION IS MENTIONED WHEN THE ESTIMATION RESULTS ARE PRESENTED Where there is some finding related to an industry, at least some attempt to explain “why” would be useful. For example, the authors say “The firms in personal services, the reference category, make up the least ‘mobile’ industry.” Why might this be the case. - THE RESULTS SECTION CONTAINS MORE INTERPRETATIONS OF WHY A RESULT IS FOUND I think a more detailed discussion of Hypothesis #2 is in order. This is pretty important and gets only two paragraphs. Similarly in Hypothesis #3 when the authors talk about industry differences I wonder if it is really the outputs that generate the difference (e.g. industry) or the inputs (say K/L ratio) that fosters the observed differences in the effect. (This would be a great follow-on paper!) - SEE PREVIOUS COMMENT In hypothesis #5 more discussion about the urban findings would also help. This is another important result. So, my criticism is that they need more not less. I worked the model, and found no errors, and see nothing in the econometric approach other than the interpretation of the magnitudes that warrants clarification. ________________________________________________________