APPENDIX H. Statistical Analysis

Transcription

1 APPENDIX H Statistical Aalysis

2

3 STATISTICAL ANALYSIS METHODS Bruce Owe INTRODUCTION The statistical aalyses i this report share a commo approach ad methodology, described i this sectio. Methodological details specific to each material type icludig aimal boe, ad ic cotaiers, ad ic tableware are described i their respective sectios. The collectios used from these aalyses are those from the Cypress Replacemet Project i West Oaklad (Oaklad), the SFOBB West Approach Project archaeological study i Sa Fracisco, ad the curret archaeological work for SF-80 Bayshore Project. The aalyses all seek statistically sigificat patterig i the distributio of geeral categories, such as beef boe or beer bottles, amog excavated features divided accordig to potetially meaigful cultural categories established from documetary sources. The artifact categories are listed i their respective sectios. The classificatio of features is summarized i the table below. Category Ethicity Icludes/ratioale Sa Fracisco: Germa, Irish, Eglish, U.S.-bor white, Mixed Oaklad: Africa America, Germa, Irish, U.S.-bor white, several ucommo ethicities. Mixed ad ucommo ethicities are icluded whe comparig each of the above with "all others." Occupatio category Dwellig type Wealthy Professioal, Professioal, Skilled, Uskilled. Households with Semiskilled workers are lumped with Skilled. Sa Fracisco: Sigle-family residece (may have boarders/lodgers); Multifamily residece (duplex, flats, may have boarders/lodgers); "Lodgigs" (multiple, presumably low-status); Commercial property with lodgig Oaklad: Differet housig categorizatio, ot cosidered here Street frotage Mai (Harriso, Bryat, Folsom); Iterior (Chelsey, Kate, Maria, Perry, Clemetia); Numbered (Seveth, Eighth, Third). Oaklad: Not available. Teacy status Ower, Teat, Ukow. Missig data are treated as Ukow.. Neighborhood Sa Fracisco: Missio Bay, Rico Hill Oaklad: East of Market Street, West of Market Street, Oaklad Poit City Feature type Sa Fracisco or Oaklad Privy, Well, Other/mixed H.1

4 H.2 SF-80 BAYSHORE ARCHAEOLOGY PROJECT The cotext data used here were provided by Mary Praetzellis i April I reported o similar aalyses for the Cypress Project i Oaklad (Owe 2004). The comparative data from that project used here is limited to cotexts datig to before 1890, ad icludes some additios. The Sa Fracisco cotexts are categorized usig a slightly differet set of variables. The methodology used here is substatially the same as that used i the Cypress aalysis, ad the descriptio below is oly slightly modified from that report, except for the additio of a discussio of miimum sample sizes. APPROACH AND METHODS Most of these aalyses use percetage data i order to look at the compositio of the assemblage from each "aalytical uit, without cosiderig the relative amouts of material from each cotext. The assumptio i usig percetage data i this way is that, over the log ru, households cosumed relatively similar amouts of meat,, ics, ad so o. Thus so comparig the proportios of specific types of boe,, ad so o relative to the total amout recovered should brig out differeces i cosumptio that would otherwise be masked by culturally uimportat differeces i the size ad artifact desity of features excavated. This is clearly a imperfect assumptio, ad it should be bore i mid that these aalyses refer to percetages of items withi their artifact classes (as i "15% of all idetifiable meat weight from feature X was beef"), ot to absolute amouts cosumed. The aalytical uits i these aalyses are sigle or multiple stratigraphic uits that are take to represet a sigle sample of refuse from a sigle residetial cotext, such as a house or a hotel. Each such cotext is represeted by oly oe aalytical uit, ad each feature is take to represet just oe residetial cotext, although this may be a simplificatio i some cases. By aalyzig the percetage compositio of artifact types (aimal boe, bottle, etc.) from each aalytical uit, differeces i the size of these features ad their depositioal history are elimiated from cosideratio. Agai, oly the mix of artifact types is cosidered here; the amouts discarded are ot evaluated. The statistics used give equal weight to each feature. I effect, each aalytical uit represets the mix of artifact types discarded by a sigle residetial uit. The aalyses are comparisos of the artifact-type mixes of these residetial uits. The aalyses proceed i steps, summarized here. 1. Select features suitable for the particular aalysis. 2. Prit a table showig the average values of each of the variables of iterest for each category (such as Wealthy Professioal, Professioal, Skilled, ad Uskilled) 3. Compare pairs of categories (such as Professioal vs. Skilled) to see if ay variable (such as beef) is sigificatly differet i the two categories of cotexts.

5 Appedix H: Statistical Aalysis H.3 4. Do similar pairwise comparisos usig lumped categories (such as Irish vs. all o-irish) 5. Do additioal pairwise comparisos betwee comparable categories i Sa Fracisco ad Oaklad (such as Professioals i Sa Fracisco vs. Professioals i Oaklad). 6. Do comparisos of categories (such as Professioals vs. Skilled) i the etire sample of Sa Fracisco ad Oaklad, lumped together as represetative of the urba Sa Fracisco Bay area. 7. Iterpret the results. 8. Perform additioal, differet statistical tests to aswer specific questios that arise. First, the features to be icluded i ay give compariso were selected to iclude oly those for which the relevat cotext data were available. Additioal restrictios were also applied i may cases, for example limitig the cases to residetial, as opposed to commercial, properties. Secod, the data were summarized accordig to the cotext variables (such as Eglish, Germa, Irish, U.S.-bor white) ad reported i a table showig the mea values of all the variables (such as pork). These values average the percetages of the features, so small features cout the same as large oes. They give a sese of the cetral tedecies of each cotext category. These meas of percetages may ot add up to 100 percet. These tables of mea values are useful exploratory tools, but they are deceptively difficult to iterpret. The mea values may hide a great deal of variatio, ad especially with the small sample sizes here, the differeces they suggest may ot be meaigful. How large must a differece be to be cosidered importat? How close must two percetages be to be cosidered effectively the same? It is eve possible for the meas to be idetical whe there is actually a real differece betwee the categories. Cosider a hypothetical case i which all the features from Latvia households had aroud 10% groomig bottles, while amog the five Estoia households, four features had o groomig bottles ad oe had 100% groomig bottles, for a average of 20%. The mea values would suggest that Estoia households typically had a higher proportio of groomig bottles tha did Latvia households, whe i fact the opposite was true. The third step of the aalysis attempts to resolve these problems by evaluatig the statistical sigificace of the differeces betwee categories of features. The statistics used are oparametric, that is, they do ot assume a ormal (bell-shaped) distributio of values. This is importat, sice the small sample sizes mea that the luck of the draw is likely to produce o-ormal sample distributios eve if the uderlyig patters are ormal. Moreover, humas are complicated, ad there is o reaso to assume ormal distributios of behavior i such historically particular, idividualistic matters as food prefereces. Parametric tests, such as the familiar t-test, will ofte fid "sigificat"

6 H.4 SF-80 BAYSHORE ARCHAEOLOGY PROJECT differeces betwee small samples of archaeological data simply because they are ot ormal ad thus fit poorly to the t-test's ull hypothesis: that samples are draw from a sigle ormal distributio. The statistic used here is the Wilcoxo rak-sum test (also called the Ma- Whitey-Wilcoxo Test) for cases with two classes (such as a compariso of percet alcohol bottles i Professioal features vs. Uskilled features). This statistic is well explaied i the followig source: Gibbos, Jea D Noparametric Statistics: A Itroductio. Sage Uiversity Papers Series o Quatitative Applicatios i the Social Scieces, Sage Publicatios, Newbury Park, Califoria. I essece, the test arrages all the values i rak order, from smallest to largest, disregardig the size of the differeces betwee them. If the percetage of alcohol bottles was greater i sigle-family houses tha i multifamily houses, the values from sigle-family houses would mostly be towards the high ed of the list, ad the values from multifamily houses would mostly be towards the low ed. If the percetage of alcohol bottles was the same i sigle-family ad multifamily houses, the the values for each kid of house would be uiformly scattered through the whole list. The tests evaluate whether or ot the list is sigificatly ubalaced by calculatig the odds of gettig a patter at least that ubalaced if you were to put the values i order by chace, such as by radomly drawig "sigle-family" or "multifamily" from a collectio of slips of paper with the appropriate umber of each type. If the chace of gettig a list as uevely distributed as the observed oe is low (less tha 10%, or less tha 5%), the the patter is deemed to be sigificat, that is, probably due ot to chace, but to a real differece betwee the two categories. The Wilcoxo rak-sum test is used to compare all the possible pairs of categories, such as sigle-family vs. multifamily houses. These results are easy to iterpret: a sigificat result meas that the variable (such as percet alcohol bottles) is sigificatly differet i the two categories. Sigificatly differet meas that the differece is cosistet eough that it is ulikely to be radom, so it is appropriate to look for a cultural explaatio. A differece with a probability of 5% has oly a 5% chace of havig occurred radomly, so we ca cosider it probably the result of some systematic process, rather tha the luck of the draw. A sigificat result does ot mea that the differece is large. A real, sigificat differece might evertheless be subtle ad ot very importat. Cosider the difficulty of iterpretig a fidig that Latvia households cosistetly used 1% more groomig bottles tha Estoia oes. Sigificat differeces idicate treds i the data that should be take seriously. The patter that appears is probably due to a real process, but the iterpretatio is up to the archaeologist. The fourth step repeats the third, but usig lumped categories such as features from professioal ad wealthy professioal households vs. skilled ad uskilled workers' households, or Irish vs. all others.

7 Appedix H: Statistical Aalysis H.5 The fifth step is logically similar to the previous two, but it compares categories i Sa Fracisco with their couterparts i Oaklad, limited to cotexts dated to before 1890 i order to improve cotemporaeity with the Sa Fracisco sample. This process starts by lumpig all the Sa Fracisco cotexts ad comparig them to all the Oaklad cotexts, ad cotiues by comparig arrower categories with the aalogous oes i the other city. This is possible oly for categories that are idetified i cities. There are o comparisos of dwellig types across the two cities, for example, because the dwellig-type categories are differet i each. Similarly, there is o compariso of Africa America households i Sa Fracisco versus Oaklad, because o Africa America households were sampled i Sa Fracisco. The sixth step is, agai, logically similar. I this case, I lump the etire sample from Sa Fracisco ad Oaklad together, treatig it as a sigle populatio represetative of the urba Sa Fracisco Bay area. The larger sample size may permit more subtle patters to be detected amog the categories (such as Germas vs. all o-germas). The seveth step is iterpretatio, i which the results are subjectively evaluated to see if they make sese. I have doe this i part by orderig the tables of sigificace tests so as to juxtapose comparisos that seem to be related, helpig me ad the reader to abstract geeralizatios from them. Others might otice ad emphasize differet patters i the results. It is also importat to look for multiple tests that cofirm related treds. This is because the method used here is iductive. That is, I did ot start with a hypothesis ad test the data to evaluate it. Istead, I ra all the reasoable comparisos I could thik of, ad pulled out for discussio those that proved sigificat either at the 10% level (less tha 10% chace that the two categories actually have idetical distributios of values, that is, less tha 10% chace that the differece is a illusio caused by the luck of the draw) or at the more covicig 5% level (less tha 5% chace that the differeces are a illusio caused by the luck of the draw). This procedure will produce some spurious "sigificat" results by chace. That is, out of oe hudred tests of two samples of a sigle distributio of values, five are expected to show differeces "sigificat" at the 5% level, just by chace. For this reaso, isolated sigificat results may or may ot reflect real cultural processes. Where multiple sigificat results seem to reflect a sigle uderlyig tred, the the tred ca be cosidered real. The eighth step is eeded oly whe the previous oe raises specific questios. For example, i the faual aalysis, some results suggested a uexpected relatioship betwee meat-type prefereces ad the cost of cuts. I order to evaluate this, I made plots of meat-type percetages versus cut-cost percetages, ad calculated correlatios betwee the variables. These additioal checks are described i their respective sectios of the report. Fially, the lack of statistically sigificat differeces betwee most of the categories does ot mea that there ecessarily are o differeces betwee the categories. It simply meas that ay differeces preset are ot great eough to be detected with cofidece based o the give sample size ad variability.

8 H.6 SF-80 BAYSHORE ARCHAEOLOGY PROJECT The statistics were ru o SAS software, usig programs ad data files specified i their respective sectios. Percetages are preseted rouded to the earest percet, but all calculatios are doe to the full precisio of the origial data. LIMITATIONS DUE TO SMALL SAMPLE SIZES The ability to detect ad iterpret quatitative patters is limited whe the sample size is small. There are two distict aspects to sample sizes: the umber of cases beig compared (for example, the umber of features from Missio Bay households ad the umber from Rico Hill households), ad the umber of artifacts that cotribute to the value of each case (for example, 5 liquor bottles out of 10 bottles total i a particular feature). First cosider the umber of cases, or features. The statistical tests used here (Wilcoxo rak-sum tests comparig two categories) have well-defied miimum sample sizes (summarized i the table below), below which they are icapable of showig statistical sigificace. There is o poit i eve cosiderig the sigificace test results if the sample sizes are below these limits, because o matter how drastically ad cosistetly the two categories might actually differ, the test is uable to show sigificace. Miimum Sample Sizes for 10% ad 5% Cofidece Usig the Wilcoxo Rak-Sum Test # i category 1 # i category 2 Total Ca show 10% sigificace Ca show 5% sigificace Yes No Yes Yes Yes No Yes Yes Yes No Yes Yes Yes No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No

9 Appedix H: Statistical Aalysis H.7 These miimum umbers of aalytical uits assume that the two categories differ so much that they do ot overlap at all. If the differeces are more subtle, the sample sizes must be larger to detect a sigificat differece. For example, say Missio Bay households averaged 5% liquor bottles out of all bottles i each feature, with a rage from 1% to 10%, while skilled households averaged 10%, with a rage from 5% to 15%. Eve if this differece were real, ad uskilled households really did ted to have fewer liquor bottles tha Rico Hill households, a few of the hardest-drikig Missio Bay households would still have a higher percetage of liquor bottles tha a few of the mostrestraied Rico Hill households. The miimum sample sizes metioed above would ot be sufficiet to detect this real differece as sigificat. The more overlap betwee the categories, the larger the size would have to be i order to detect a sigificat differece. These miimum sample-size requiremets are modest, eve compared to other tests that require more striget assumptios about the data. The Sa Fracisco dataset, however, is small eough that may ways of subdividig it fall below these sample-size thresholds. This meas that o matter how real ad strog the patters may be, i these cases i which the sample sizes fall below the miima, they will ot be detected. By the same toke, this meas that absece of sigificat differeces does ot ecessarily imply that the households i the two categories are similar. The reader ca apply these miimum sample sizes to the artifact aalyses by cosiderig the "" values give i the percetage data tables. These values idicate how may features are icluded i each category. For example, there are 2 uskilled households ad 3 professioal households i the Sa Fracisco sample. I effect, there are o statistical results comparig uskilled ad professioal households, because the sample sizes are below the miimum required to show 10% sigificace. We simply caot determie if there are ay sigificat differeces betwee these categories of households, ad the absece of ay such differeces is i o way evidece that they were similar. The other aspect of sample size is the umber of artifacts that are used to calculate the value for each feature. Say two pickle jars are foud i the etire project, ad they happeed to be from two of the three sigle-family households tested. Eve if the sample of features is large eough to show that sigle-family households have sigificatly more pickle jars tha multifamily households, how seriously should we take this result? While the sigificace test shows that this distributio of pickle jars is ulikely to have happeed by chace, the percetages of pickle jars i each feature are determied by a very small umber of items. Pickle jars are sufficietly ucommo that oe or two idiosycratic purchases could accout for the etire patter. These may have little real meaig i terms of geeral differeces betwee kids of households. If the very same patter were established o the basis of 100 pickle jars, it might be o more sigificat i the statistical sese. The umber of sigle-family household features with higher percetages of pickle jars might be exactly the same. But with a larger umber of artifacts cotributig to each of those percetages, we would be justified i thikig that

10 H.8 SF-80 BAYSHORE ARCHAEOLOGY PROJECT the differece betwee features reflected a cosistet behavioral patter, ot just a few isolated evets. This aspect of sample size is ot quatifiable. I the pickle-jar example, we might require well over two jars i order to be coviced of a patter. But if the project foud oly two gold rigs, ad were from wealthy professioals' households, we would rightly be more willig to believe that they reflected a geeral differece betwee wealthy professioal households ad others, because the meaig of the artifacts ad the patter is more obvious. Sice I did ot have access to the raw faual data, I could ot evaluate this aspect of sample size for the meat aalysis. I the cotaier ad ic aalyses, I have oted results that are based o very small umbers of items. I these cases, eve whe the differeces are deemed sigificat, their iterpretatio should be subject to careful, culturally iformed scrutiy.

11 Appedix H: Statistical Aalysis H.9 STATISTICAL ANALYSIS OF MEAT WEIGHT PERCENTAGES Bruce Owe SUMMARY OF FINDINGS The most strikig aspect of the Sa Fracisco meat cosumptio data is its cosistet ad frequetly statistically sigificat cotrast with comparable data from pre-1890 Oaklad recovered by the Cypress Project. Households i Sa Fracisco cosumed a mix of meats that was higher i beef ad pork, lower i mutto, ad probably lower i chicke, tha did households i Oaklad. Oaklad households also cosistetly purchased more medium-cost meat tha did Sa Fracisco households, although the meaig of this differece is hard to discer. These prefereces cross-cut most or all aalyzed groups withi each city, geerally affectig all households similarly, regardless of professio category, ethicity, homeowership or teat status, ad eighborhood. This city-wide, cross-cuttig patterig suggests that the type prefereces, ad possibly the puzzlig preferece for medium-cost cuts, may have bee iflueced by large-scale factors that would have affected each city as a whole, such as the city's locatio relative to differet meat-producig areas or i the etwork of shippig or meat-distributio routes. There is little statistically sigificat patterig by professio, ethicity, dwellig type, home owership, or type of street i the summary faual data from the SF-80 ad WBA projects. This is probably due i part to the small sample sizes, but may also idicate that these variables did ot have a very strog effect o people's meat diet. The pricipal patterig i the meat data withi Sa Fracisco is by eighborhood. The residets of the Rico Hill eighborhood cosumed the most expesive mix of cuts of meat of ay eighborhood i either city. People i the three Oaklad eighborhoods cosumed a slightly cheaper mix of cuts, with Oaklad Poit residets cosumig the cheapest mix i Oaklad. Fially, the residets of the Missio Bay eighborhood ate a dramatically cheaper mix tha people i ay of the other eighborhoods. Nevertheless, residets of the apparetly more cost-coscious Missio Bay eighborhood cosumed a higher proportio of beef tha did people i ay of the Oaklad eighborhoods, although this was sigificat oly with respect to the two more prosperous Oaklad eighborhoods East ad West of Market. This probably reflects the Sa Fracisca city-level preferece for, or better access to, beef. It also suggests that a tedecy to purchase beef may ot have bee directly tied to ecoomic stadig, sice the eighborhoods with the cheapest ad the most expesive mixes of meat cuts cosumed relatively high proportios of beef. People i the pre-1890 West of Market eighborhood of Oaklad ate a higher proportio of mutto tha did people i ay of the other eighborhoods. I the Cypress aalysis, I suggested that this patter might be due to the higher percetage of Irish

12 H.10 SF-80 BAYSHORE ARCHAEOLOGY PROJECT residets i the West of Market eighborhood. This hypothesis is weakeed by results from Missio Bay, which was eve more heavily Irish, yet had sigificatly lower proportios of mutto tha did West of Market. Moreover, there is o sigificat differece i the proportio of mutto betwee Irish ad kow o-irish households i Sa Fracisco or i cities lumped together. Irish ethicity is probably ot a importat factor i the differig prefereces for mutto. Comparisos with the two Sa Fracisco eighborhoods highlighted differeces i the Oaklad eighborhoods. The East of Market eighborhood was probably the closest parallel i the Oaklad sample to the apparetly prosperous Rico Hill i Sa Fracisco. Coversely, fidigs support other idicators that the Oaklad Poit eighborhood was the most ecoomically limited of the Oaklad eighborhoods. Cotrary to tetative fidigs i the Cypress aalysis, beef was ot clearly the most valued meat type, sice it did ot correlate positively with higher proportios of expesive cuts, or did it compare egatively with proportios of cheap cuts. I Sa Fracisco oly, mutto correlated with more expesive mixes of meat cuts, suggestig that mutto was more highly valued i Sa Fracisco tha i Oaklad. I Oaklad, pork was probably the least preferred meat type by this criterio, ad there is a hit of a similar patter i Sa Fracisco. The differeces betwee cities i the relative valuatio of meat type further reiforces the coclusio that decisios cocerig meat type ofte reflected cosideratios other tha the ecoomic oes that affected the choice betwee cheap, medium, or costly cuts, ad that some of these cosideratios were oes that affected each city as a whole. No sigificat differeces were detected betwee privies ad wells withi Sa Fracisco, but lumpig the features from cities idicated that the three wells i the sample of residetial households accumulated more high-cost cuts tha did privies, albeit with oly 10% cofidece. This patter is ot reassurig, sice it suggest that privies ad wells might ot be comparable samples of household behavior, ad if ot, that they should ot be treated as equivalet i the same aalysis. Two of the wells, however, were located i the apparetly privileged Rico Hill eighborhood, so the differeces may simply reflect the high ecoomic status of the cosumers that the wells represet. Alteratively, with such a small sample of wells, the patter could be due to chace, or to the uusual behavior of just oe or two households with wells. O the assumptio that oe of these alteratives ca probably explai away the seemig differece betwee the two types of features, ad i the iterest of ot further reducig the already small sample sizes, the wells are left i the preset aalysis. Future projects should be attetive to the possibility that wells ad privies might ot produce equivalet samples for aalysis. Further aalyses, icludig a multivariate approach, might be helpful i disetaglig the effects of the differet classificatio variables here.

13 Appedix H: Statistical Aalysis H.11 INTRODUCTION This report describes results of a search for statistically sigificat patterig i the distributio of meat type (beef, mutto, pork, ad chicke) ad meat-cut cost categories (high, medium, ad low amog the three mammal meats) amog features divided accordig to potetially meaigful cultural categories. The meat is measured i estimated meat weight for the type or cost of cut, as a percetage of total meat weight estimated for the feature (see Food-refuse Aalysis: Faual Remais i Chapter 6). Oly beef, mutto, pork, ad chicke are icluded i the aalysis; meat weight for other birds, rabbit, ad so o is igored. The cultural categories are the same as those used i the other statistical aalyses. They are summarized i the "Statistical aalysis methods" sectio. The faual data i the form of percetages of meat weight were provided by Mary Praetzellis i early March The cultural cotext data were also provided by Mary Praetzellis, i April The comparative data from Oaklad are from the Cypress Project. For this aalysis, these data are limited to cotexts datig to before 1890, ad they iclude some additios sice the Cypress report was prepared. APPROACH AND METHODS This aalysis follows the geeral methodology described i uder Statistical Aalysis Methods, above. Some results suggested a uexpected relatioship betwee meat-type prefereces ad the cost of cuts. I order to evaluate this, I made plots of meat-type percetages versus cut-cost percetages, ad calculated correlatios betwee the variables. These additioal checks are described i the detailed sectio of the report. As oted i the methods sectio, sigificat patters based o small umbers of items should be assessed with cautio, sice a few idiosycratic idividuals or actios might accout for the patter, rather tha ay broad tedecies of the household categories beig compared. Sice the data for this aalysis were already coverted to percetages, however, it is impossible to evaluate how may boes, aimals, or purchases may make up ay give sample. A total of 100 idetifiable boes was required to be icluded i the faual aalysis, this was deemed to be a sufficietly large umber of meat purchases to be take as represetative of cosistet behaviors by the members of each household. The statistics were ru o SAS software, usig SAS istructios i the program CSMEAT2.SAS, faual data from CYSFMEA2.DBF, ad cotext data from CSCTX4.DBF. The program is a simple text file that ca be viewed usig ay word processor, ad the data files ca be viewed directly by Excel or most database programs.

14 H.12 SF-80 BAYSHORE ARCHAEOLOGY PROJECT RESULTS: WITHIN SAN FRANCISCO ONLY OCCUPATION (SF-80 AND WBA ONLY) Occupatio Category Number of Features Percet Beef Percet Mutto Percet Pork Percet Meat-cut Costs Percet Chicke High Medium Low Wealthy Professioal Professioal Skilled Uskilled Total: Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: All pairs of occupatio categories Wealthy (P+) vs. all others (P,S,U) Wealthy (P+) vs. Middle (P,S) Middle (P,S) vs. Uskilled (U) Middle (P,S) vs. Extremes (U,P+) Upper (P+,P) vs. Lower (S, U) Ay skill (P+,P,S) vs. Uskilled (U) Sigificat Differeces (comparisos that reached at least 10% sigificace): Occupatio Comparisos: Variable Which has more? Probability 5% 10% Professioal vs. Skilled Pork Professioal X Skilled vs. Uskilled Pork Uskilled X Professioal vs. Skilled Chicke Skilled X Iterpretatios: There are o cosistet ad covicig patters i meat cosumptio by occupatio category withi the Sa Fracisco sample. Two of the three sigificat differeces idicate that more pork was cosumed by the households of higher ad lower professioal status tha by those of skilled workers. The percetage data are o less cotradictory, ot suggestig ay itelligible treds for ay type or for costs of cuts.

15 Appedix H: Statistical Aalysis H.13 The sample sizes for each professio category are so small that some comparisos caot produce sigificat results o matter how strogly the groups differ (see the discussio i the methods sectio), while the remaider is sesitive oly to very strog differeces. By the same toke, the use of such small samples meas that oe or two strogly idiosycratic households may create sigificat patters, which is probably the case here. The two professioal households average relatively extreme prefereces for pork, agaist chicke ad mutto, ad agaist expesive cuts, cosiderig their assumed ecoomic stadig. It would probably be uwise to ifer aythig about geeral tedecies of behavior from these data, other tha that these two professioal households may be uusual i some way. ETHNICITY (SF-80 AND WBA ONLY) Ethicity Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Eglish Germa Irish U.S.-bor white Total: Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: All pairs of commo ethicities Each of the four commo ethicities vs. the other three lumped together Each of the four commo ethicities vs. all the others lumped together, icludig mixed oes (ot listed i the table above) No sigificat differeces were foud. Iterpretatios: There is o sigificat patterig i meat cosumptio by ethicity withi the Sa Fracisco sample. This may be due to the small sample sizes for each ethicity, which make the tests sesitive oly to very proouced differeces.

16 H.14 SF-80 BAYSHORE ARCHAEOLOGY PROJECT DWELLING TYPE (SF-80 AND WBA ONLY) Dwellig Type Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Commercial with lodgig Lodgigs, multiple Multifamily housig Sigle-family residece Total: Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: All pairs Sigle-family vs. all others except commercial Sigle-family vs. all others, icludig commercial No sigificat differeces were foud. Iterpretatios: There is o sigificat patterig i meat cosumptio by dwellig type withi the Sa Fracisco sample. Agai, this may be due i part to the small sample sizes for each dwellig type, which make the tests sesitive oly to proouced differeces. TENANCY (SF-80 AND WBA ONLY) Teure type Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Ukow Ower Reter (Teat) Total: Table excludes commercial property. Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos.

17 Appedix H: Statistical Aalysis H.15 Comparisos: Ower vs. Reter (Teat) No sigificat differeces were foud. Iterpretatios: There is o sigificat patterig i meat cosumptio by teacy withi the Sa Fracisco sample. Agai, this may be due i part to the small sample sizes, which make the tests sesitive oly to proouced differeces. NEIGHBORHOOD (SF-80 AND WBA ONLY) Neighborhood Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Missio Bay Rico Hill Total: Table icludes commercial properties. Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: Missio Bay vs. Rico Hill Sigificat Differeces (comparisos that reached at least 10% sigificace): Neighborhood Variable Which has more? Probability 5% 10% Missio Bay vs. Rico high cost Rico Hill X X Missio Bay vs. Rico low cost Missio Bay X X Missio Bay vs. Rico mutto Rico Hill X Iterpretatios: The mix of mammal meat cuts foud i the Rico Hill eighborhood had a sigificatly higher proportio of high-cost cuts, ad a sigificatly lower proportio of low-cost cuts, tha the mix foud i the Missio Bay eighborhood. Both of these patters are sigificat at the 5% cofidece level, ad they ted to cofirm each other. This is ot automatically the case, sice depedig o the proportio of medium-cost cuts, either oe could be sigificat without the other, or oe extreme could be sigificat alog with a sigificat differece i the medium-cost cuts. The meat-cut data suggest that people i Rico Hill teded to cosume more expesive cuts of meat tha those i Missio Bay.

18 H.16 SF-80 BAYSHORE ARCHAEOLOGY PROJECT People i Rico Hill probably also cosumed a higher proportio of mutto tha did those i Missio Bay, although this tred is sigificat oly at the 10% level. Sice Rico Hill residets preferred a costlier mix of meats, this suggests that mutto may also have bee relatively prized. Results cocerig type prefereces are complicated, ad are discussed i other sectios below. STREET FRONTAGE TYPE (SF-80 AND WBA ONLY) Street Frotage Type Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Iterior Mai Numbered Total: Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: All pairs. No sigificat differeces were foud. Iterpretatios: There is o sigificat patterig i meat cosumptio by street frotage withi the Sa Fracisco sample. Agai, this may be due i part to the small sample sizes, which make the tests sesitive oly to proouced differeces. FEATURE TYPE (SF-80 AND WBA ONLY) Feature type Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Other/ combied Privy Well Total: Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos.

19 Appedix H: Statistical Aalysis H.17 Comparisos: Privy vs. Well No sigificat differece was foud. Iterpretatios: There is o sigificat differece i the meat assemblages from privies ad wells withi the Sa Fracisco sample. This fidig suggests that icludig types of cotexts i this aalysis probably does ot itroduce serious biases. RESULTS: COMPARISONS BETWEEN SAN FRANCISCO AND OAKLAND PRE-1890 CONTEXTS ALL SAN FRANCISCO CONTEXTS VS. ALL OAKLAND CONTEXTS City Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Sa Fracisco Oaklad Total: Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: All Sa Fracisco vs. all Oaklad Sigificat Differeces (comparisos that reached at least 10% sigificace): City Variable Which has more? Probability 5% 10% SF vs. Oaklad Beef SF X SF vs. Oaklad Pork SF X SF vs. Oaklad Mutto Oaklad X X SF vs. Oaklad Medium Oaklad X X Iterpretatios: There seems to have bee a sigificat differece i the overall mix of meats cosumed i Sa Fracisco compared to Oaklad. Households i Sa Fracisco probably cosumed a mix of meats relatively higher i beef ad pork, although patters are idicated oly at the 10% cofidece level. Coversely, ad eve more clearly idicated at the 5% level, people i Oaklad cosumed more mutto tha did

20 H.18 SF-80 BAYSHORE ARCHAEOLOGY PROJECT those i Sa Fracisco. The percetage data suggest that people i Oaklad may have cosumed more chicke, too, but the differece is ot sigificat. Refuse from Oaklad has a relatively higher proportio of medium-cost cuts, but the iterpretatio of this differece is ot clear. With o corroboratio from other cost comparisos, this patter ca probably be disregarded as ot very meaigful, oe of the spurious results expected by chace, or. OCCUPATION IN SAN FRANCISCO VS. THEIR COUNTERPARTS IN OAKLAND City Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Sa Fracisco P Oaklad P Total P Sa Fracisco P Oaklad P Total P Sa Fracisco S Oaklad S Total S Sa Fracisco U Oaklad U Total U Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: Sa Fracisco P+ vs. Oaklad P+ Sa Fracisco P vs. Oaklad P Sa Fracisco S vs. Oaklad S Sa Fracisco U vs. Oaklad U

21 Appedix H: Statistical Aalysis H.19 Sigificat differeces (comparisos that reached at least 10% sigificace): City Variable Which has more? Probability 5% 10% SF Skilled vs. Oak Skilled beef SF Skilled X SF Uskilled vs. Oak Usk pork SF Uskilled X SF Skilled vs. Oak Skilled mutto Oaklad Skilled X SF Uskilled vs. Oak Usk mutto Oaklad Uskilled X SF Prof vs. Oaklad Prof chicke Oaklad Prof X SF Skilled vs. Oak Skilled high SF Skilled X SF Skilled vs. Oak Skilled medium Oaklad Skilled X Iterpretatios: This sectio repeats the same comparisos betwee cities as i the precedig sectio, but limits them to a sigle occupatio category i each city. This procedure reduces the sample sizes compared to the whole-city aalysis, makig the tests less sesitive, but it also reduces possible cofoudig effects that might cofuse the picture if some of the betwee-city differeces were expressed differetly amog differet occupatioal variables. For example, if the lifestyle of wealthy professioals i Sa Fracisco differed from that of wealthy professioals i Oaklad, while uskilled laborers lived similarly i cities, this aalysis should brig those patters out. I fact, these comparisos suggest that i geeral, the whole-city treds oted above may apply across most or all of the occupatioal categories. That is, skilled households i Sa Fracisco cosumed more beef tha did their couterparts i Oaklad; uskilled households i Sa Fracisco cosumed more pork tha did their couterparts i Oaklad; ad skilled ad uskilled households i Oaklad cosumed more mutto tha did comparable households i Sa Fracisco. These are precisely the same type prefereces that distiguished the two cities as a whole. Moreover, the percetage data suggest that although the patters are ot sigificat, all three city-level type prefereces also may have held amog professioal households, ad that amog wealthy professioals, the Sa Fraciscas probably reflected their city's preferece for beef, although they did ot differ clearly from their Oaklad couterparts o pork or mutto. I other words, the Sa Fracisca preferece for beef ad pork affected people i all occupatioal categories, as did the Oaklad preferece for mutto. Oaklad professioal households cosumed more chicke tha did professioal households i Sa Fracisco. The percetage data suggest a similar, but ot sigificat, tedecy of Oaklad wealthy professioals ad skilled workers to prefer chicke more tha did their couterparts i Sa Fracisco. Uskilled workers deviate from this patter, but ot by much. Fially, skilled households reflect the same puzzlig city-level patter i which Oakladers cosumed a higher proportio of medium-cost cuts of meat. I the aalysis

22 H.20 SF-80 BAYSHORE ARCHAEOLOGY PROJECT limited to just skilled households, this patter is matched by a complemetary oe i which Sa Fracisca skilled workers cosumed a correspodigly higher proportio of expesive cuts. The percetage data o meat-cut costs amog the professioal categories i the two cities do ot suggest ay cosistet or itelligible patter. The meat type fidigs reiforce the impressio of city-level differeces i meat type prefereces that apparetly cross-cut professioal categories. This suggests that these differeces are probably to be explaied by processes that would have affected people of all professios i the same way, but were distict i each city, such as the locatio of the city relative to meat producers ad i trasportatio ad trade etworks. OWNERS AND TENANTS IN SAN FRANCISCO VS. THEIR COUNTERPARTS IN OAKLAND City Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Sa Fracisco teats Oaklad teats Total teats: Sa Fracisco owers Oaklad owers Total owers: Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: Owers i Sa Fracisco vs. Owers i Oaklad Teats i Sa Fracisco vs. Teats i Oaklad No sigificat differeces were foud. Iterpretatios: Although either owers or reters differed sigificatly from oe city to the other, the percetage data follow all of the suggested city-level differeces i meat-type prefereces ad cut costs amog teats ad homeowers i the two cities. While this is ot a statistically sigificat result, it does suggest that these patters affected

23 Appedix H: Statistical Aalysis H.21 owers ad reters, supportig the impressio that they were probably due to citylevel processes, perhaps geographic i ature. ETHNICITIES IN SAN FRANCISCO VS. THEIR COUNTERPARTS IN OAKLAND City Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Sa Fracisco Germa Oaklad Germa Total Germa Sa Fracisco Irish Oaklad Irish Total Irish Sa Fracisco U.S.-bor white Oaklad U.S.-bor white Total U.S.-bor white Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: Sa Fracisco Germa vs. Oaklad Germa Sa Fracisco Irish vs. Oaklad Irish Sa Fracisco U.S.-bor white vs. Oaklad U.S.-bor white Sigificat Differeces (comparisos with at least 10% sigificace): City Variable Which has more? Probability 5% 10% SF US vs. Oaklad US beef S.F. U.S.-bor white X SF Irish vs. Oaklad Irish mutto Oaklad Irish X X SF US vs. Oaklad US mutto Oaklad U.S.-bor white X

24 H.22 SF-80 BAYSHORE ARCHAEOLOGY PROJECT Iterpretatios: Oce agai, the city-level patters i meat prefereces are reflected i subsets of the cities' populatios, i this case subsets by ethicity. U.S.-bor whites i Sa Fracisco ate sigificatly more beef tha did their couterparts i Oaklad, ad Irish ad U.S.-bor whites i Oaklad ate more mutto tha did their couterparts i Sa Fracisco. The percetage data suggest that these patters geerally held for almost all the type prefereces amog all three groups, eve where the patters are ot sigificat. There is oly oe violatio of the city-level patter, amog the Sa Fracisco Irish who ate relatively more, rather tha less, chicke tha did their Oaklad couterparts. Germas followed the city-level patter for beef, ad U.S.-bor whites for pork, but by such small margis that these cases should ot be take as support for the city-level patters. Nevertheless, the overall patter suggests that the city-level meattype prefereces geerally cross-cut the ethic idetities withi each city. NEIGHBORHOODS IN SAN FRANCISCO VS. NEIGHBORHOODS IN OAKLAND Neighborhood Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Missio Bay - SF Rico Hill - SF Sa Fracisco, all cotexts East of Market - Oaklad West of Market - Oaklad Oaklad Poit - Oaklad Oaklad, all cotexts Table icludes commercial properties. Except for the umber of features (), all figures are percetages, ad all totals are weighted averages to accout for assemblage-size differeces. Chicke is reported with a additioal decimal place to facilitate comparisos. Comparisos: All pairs of eighborhoods

25 Appedix H: Statistical Aalysis H.23 Sigificat Differeces (comparisos that reached at least 10% sigificace): Neighborhood Variable Which has more? Probability 5% 10% Missio Bay SF vs. E Oak beef Missio Bay SF X X Missio Bay SF vs. E Oak mutto East of Mkt Oak X X Missio Bay SF vs. E Oak high cost East of Mkt Oak X X Missio Bay SF vs. E Oak low cost Missio Bay SF X X Missio Bay SF vs. W Oak beef Missio Bay SF X X Missio Bay SF vs. W Oak mutto West of Mkt Oak X X Missio Bay SF vs. W Oak low cost Missio Bay SF X X Missio Bay SF vs. Oak Pt mutto Oaklad Pt, Oak X X Missio Bay SF vs. Oak Pt medium Oaklad Pt, Oak X Missio Bay SF vs. Oak Pt low cost Missio Bay SF X X Rico Hill SF vs. W Oak mutto West of Mkt Oak X X Rico Hill SF vs. Oak Pt high cost Rico Hill X X Rico Hill SF vs. Oak Pt medium Oaklad Pt, Oak X Withi Sa Fracisco: Missio Bay vs. Rico high cost Rico Hill X X Missio Bay vs. Rico low cost Missio Bay X X Missio Bay vs. Rico mutto Rico Hill X Withi Oaklad: West Oak vs. E Oak mutto West of Mkt Oak X X West Oak vs. Oak Pt mutto West of Mkt Oak X X All SF vs. all Oaklad: SF vs. Oaklad beef SF X SF vs. Oaklad mutto Oaklad X X SF vs. Oaklad pork SF X SF vs. Oaklad medium Oaklad X X

26 H.24 SF-80 BAYSHORE ARCHAEOLOGY PROJECT The table below summarizes the comparisos betwee eighborhoods: Compariso Beef Mutto Pork Chicke Meat-cut Costs High Medium Low Missio Bay vs. East of Market Missio Bay vs. West of Market Missio Bay vs. Oaklad Poit Missio Bay vs. Rico Hill Rico Hill vs. East of Market Rico Hill vs. West of Market -- Rico Hill vs. Oaklad Poit ++ - West of Market vs. East of Market ++ West of Market vs. Oaklad Poit ++ East of Market vs. Oaklad Poit All Oaklad vs. all Sa Fracisco idicates that the first eighborhood has more at the 5% cofidece level -- idicates that the first eighborhood has less at the 5% cofidece level - idicates that the first eighborhood has less at the 10% cofidece level Iterpretatios: The Rico Hill ad Missio Bay eighborhoods fall at opposite ecoomic extremes i the cost of meat cuts cosumed, as show i the reordered table of mea values below: Neighborhood Features Percet Beef Percet Mutto Percet Pork Percet Percet Meat-cut Costs Chicke High Medium Low Rico Hill - SF East of Market - Oaklad West of Market - Oaklad Oaklad Poit - Oaklad Missio Bay - SF All of the Oaklad eighborhoods fall i betwee these two extremes, clusterig closer to the Rico Hill cost mix, while Missio Bay had a markedly cheaper mix of cuts tha ay of the others. Rico Hill has the highest mea percetage of expesive cuts of ay of the eighborhoods, ad Missio Bay has the lowest. Coversely, Rico