Weathering of painted wood construction:Facade restoration

Michael O. Hunt✳

A. James O’MalleyWilliam C. Feist✳

George P. McCabeJoel W. Evans II

Daniel L. Cassens✳

Until recently the preservation andrestoration of the nation’s historicallyand architecturally significant buildingshave been a worthy but mostly emotionalmatter of maintaining a bond to Amer-ica’s heritage. Now more tangible fac-tors, couched in economics, are beingsuggested for encouraging the interest inpreservation and restoration, which is in-creasing from coast-to-coast. For exam-ple, investing in existing older neighbor-hoods reduces taxpayer-funded costs forextending municipal infrastructure tooutlying suburban sprawl. It has alsobeen noted that restoration and preserva-tion of structures in a community’s olderneighborhoods is in fact a sound environ-

mental practice of recycling buildingmaterials and resources.

Historically, because of its availabilityand ease of use, wood was the prevalentconstruction material for both interior

and exterior applications. Therefore,much historic restoration and preserva-tion activity deals with economicallysolving technical problems dealing withthe use, reuse, and care of wood.

But some restoration projects withhigh visibility have had unwanted re-sults. Figure 1 shows a downtown store-front facade in Lafayette, Indiana, lessthan 5 years after restoration under theauspices and financial incentive of TheMain Street Program for Facade Reno-vation of Downtown Buildings (City ofLafayette [Indiana] 2002). Many of theprogram’s restoration projects involvenew wood materials, as do many com-mercial projects where historic store-fronts are returned to their original ap-pearance. Too frequently an outcome asdepicted in Figure 1 results, causingnegative consequences: 1) it reflects

FOREST PRODUCTS JOURNAL Vol. 53, No. 4 51

The authors are, respectively, Professor of Wood Science, Wood Research Laboratory,Purdue Univ., West Lafayette, IN 47907-2033; Assistant Professor of Statistics, Dept. ofHealth Care Policy, Harvard Medical School, Boston, MA 02115-5899; Information Special-ist, Exterior Wood Finishing, Middleton, WI 53562-1171; Professor of Statistics, Dept of Sta-tistics, Purdue Univ., West Lafayette, IN 47907-2067; Programmer/Analyst II, Dept. of Math-ematics and Physics, Univ. of Louisville, KY 40292; Professor of Wood Products, WoodResearch Laboratory, Purdue Univ., West Lafayette, IN 47907-2033. This is Paper No.16512of the Purdue Agri. Programs. The authors gratefully acknowledge Robert R. Leavitt for hisexpert inspection of the exposure panels in heat and snow and all weather conditions in be-tween. This paper was received for publication in April 2001. Article No. 9305.✳Forest Products Society Member.©Forest Products Society 2003.

Forest Prod. J. 53(4):51-60.

AbstractTodetermine thenaturalweathering resistanceofdifferentpaint formulations,paint

manufacturers coat thin, flat specimens of homogeneous substrate with their products.The essentially two-dimensional specimens are attached to paint fences and theirweatheringperformance isobservedover time. Incontrast, this researchusedcomplete,three-dimensional construction assemblages where paint type is but one of five con-struction variables, all of which potentially affect the long-term performance of thepainted surface. The other variables are surface pretreatment, type of wood, design ofthe assemblage, and types of sealant/caulk for the joints. This paper describes the struc-ture and methodology of the weathering of painted wood construction research projectand theuseof thisnovelprocess to report results for3yearsofweatheringexposure.Theresearch is continuing.

poorly on the building owner and busi-ness housed within; 2) it creates a nega-tive image of wood use in construction;and 3) it is a convincing deterrent to thepromotion of appropriate restoration ofhistoric buildings.

To produce satisfactory results, it iswell known (Williams et al. 1996) thatsurface preparation is the most impor-tant step in painting new wood or repain-ting previously painted surfaces. To re-store old wood-sided buildings (Fig. 2),it is often necessary to carefully removethick alligatored build-ups of paintdown to bare wood before repainting. If

fresh paint is applied without removingthis build-up, costly and early paint fail-ure is to be expected.

The goal of this research is to developprocedures and practices to help preventpremature failure of restored wood stru-ctures with special emphasis on restoredwood street facades, i.e., to prevent thekind of damage shown in Figure 1. Thegeneral research objectives of the pro-ject are to 1) develop a comprehensiveconstruction specification that is techni-cally and economically feasible and willresult in durable wood construction; and2) provide time-related rates of weather-

ing performance/deterioration of paint-ed wood constructions. All new woodwas used in the construction of test spec-imens. Therefore, results of this researchare applicable to both new constructionand restoration of buildings.

In this paper, the objective is to intro-duce the experimental methodology andstatistical analysis basis for this long-term project. The first 3 years of weath-ering data are used to illustrate the pro-cedure. The research is ongoing. At thistime, it is impossible to predict whenand how the panels will fail. Indeed, thedefinition of failure is evolving asweathering exposure progresses. How-ever, tentative recommendations arepossible and they are included in theDiscussion section.

Test materials and methodsPanels, specially constructed in 1996,

have been fully exposed to naturalweathering on the campus of PurdueUniversity in West Lafayette, Indiana,and have been inspected every 6 months.The test specimen, hereinafter referredto as an exposure panel, is a 5-3/8-inch-thick by 17- by 24-inch version ofactual commercial facade panels. Likethe typical actual panel, the exposurepanels present a raised center panel ef-fect. Three different designs were used.They differ in the number and length ofjoints. Since joints are potential entrypoints of water into the panel, the designaffects not only initial construction costbut also is likely related to long-termperformance. The three panel designsare shown in Figure 3.

The exposure panel for all three fron-tal designs is a three-dimensional box.The weathering performance of the ex-posed face of this box is the subject ofthis research. Care was taken in con-structing the exposure panels to ensurethat the performance of the exposed faceis not affected by conditions on thebackside of the face. The face is attach-ed to a frame of CCA-treated (chromat-ed copper arsenate) 2 by 4 southern pinelumber. This creates a stud wall-likestructure, the back of which is closedwith a 1/2-inch-thick CCA-treated panelof exterior grade southern pine plywood.To protect against unplanned incursionof water in the “stud-wall cavity,” ventedair holes and drain holes are included inthe construction. Therefore, the atmo-sphere within the exposure panel/box is

52 APRIL 2003

Figure 1. — Photo of historic building with deteriorated street facade.

Figure 2. — Historic house being prepared for painting.

maintained at the prevailing atmosphe-ric condition.

To create the exposed face, first, a3/8-inch-thick piece of medium densityoverlay plywood (MDO) was nailed tothe lumber frame. Next, the solid lumberfacing/surface parts were nailed to theMDO according to the specifications ofthe particular design. The three designsare distinguishable by the following dif-ferences in construction and materials.Design 1 has a center raised panel of

solid wood with machined convex edgesand standard cove trim molding of west-ern pine positioned next to the squareedges of the face parts to effect a con-cave machined edge appearance forthese parts. Design 2 creates the raisedcenter panel effect using back-to-backplacement of standard cove trim mold-ing in a rectangular configuration andthe face parts are presented with squareedges. The joints between solid faceparts of Designs 1 and 2 are square

edged. Lastly, Design 3 has all edges ofthe face parts mated with matching con-vex and concave quarter-round cuts; itsraised center panel is of solid wood withmachined convex edges.

All materials used in constructing theexposure panels were commercial prod-ucts, selected because of their availabil-ity. Sealant selections were based on theconstruction experience of the seniorauthor. The paint and water-repellentpreservative selections were based onprior favorable research experience atthe USDA Forest Service Forest Prod-ucts Laboratory (Feist 1990). Detaileddescriptions of materials, hardware, andfabrication methods are available in aPurdue University Laboratory docu-ment (Wood Research Laboratory1996a).

The construction specifics for thethree panel designs were developed inconsultation with a local contractor sothat subsequently full-scale versions ofthem could be readily produced com-mercially. Considering materials, mate-rial preparation and assembly times, andequipment costs, Design 3 is the mostexpensive to produce; the costs of De-signs 1 and 2 are about the same.

Design of the experiment

This experiment was designed to in-vestigate the effects of seven variables,called factors, on the weathering of thewood panels: 1) panel wood: yellow-poplar combined with western redcedar;air-dried CCA-treated southern pine;and wet CCA-treated southern pine; 2)panel design: see Figure 3 for the threedesigns and a previous reference (WoodResearch Laboratory 1996a) for con-struction specifications; 3) surface wa-ter-repellent preservative pretreatment:solvent-based and water-based; 4) paintsystem: alkyd primer and two acrylic la-tex top coats, and acrylic latex primerand two acrylic latex top coats; 5) seal-ant: applied or not applied; 6) paintcolor: white and medium gray; 7) loca-tion: 20 to 30 different locations on thepanel (Fig. 4).

For each combination of the first 5factors, 3 panels were constructed, giv-ing a total of 216 panels. The last twofactors varied within each panel. Thetwo colors were randomly assigned tothe two sides of the panel and the loca-tions are common to each panel design.

FOREST PRODUCTS JOURNAL Vol. 53, No. 4 53

Figure 3. — Close-up of test panels on exposure fence.

Figure 4. — Inspection template sheet.

Each panel was constructed primarilyfrom a mixture of yellow-poplar andwestern redcedar, or exclusively fromdry CCA or wet CCA. This led to thethree-level factor called “panel wood,”which indicated whether the panel wasconstructed from yellow-poplar/westernredcedar, dry CCA, or wet CCA. Be-cause the performance of yellow-poplarand western redcedar was expected to bedrastically different for some outcomes,a secondary analysis, which distin-guished locations within panels whereyellow-poplar was used from those in

which western redcedar was used, wasalso performed. The factor name“wood” was used in place of “panelwood” for such analyses.

A one-part polyurethane sealant(specifications given in Table 1) wasapplied to 36 additional panels of thesecond panel design, 3 for each combi-nation of panel wood, surface pretreat-ment, and paint type. In addition, 30panels were prepared with no surfacepretreatment, 5 for each combination ofpanel wood and paint type, all of the firstpanel design.

Paint appliedThe amount by weight of paint ap-

plied by brush to the panels for each coat(one primer coat and two top coats) wasrecorded. The film thickness of the totalamount of paint applied in the threecoats is related to the performance of thepanel and could be calculated fromweight applied. This variable is includedas a covariate in all analyses.

Exposure fenceNatural weathering is the testing agent

in this experiment. A fence made ofCCA-treated southern pine posts and 2-by 4-inch lumber was erected in a securefenced area on the west edge of thePurdue University campus. The univer-sity is located in West Lafayette, Indi-ana, which is at 40º 20’ N and 86º 55’ Wand 60 feet above sea level. Each panelwas tightly fastened to a randomly as-signed position on the fence. There aretwo tiers of panels on the fence; the up-per tier is vertically offset from the bot-tom so that drainage from the top tierdoes not influence the weathering of thepanels on the bottom tier. All panels arehung vertically, as if they were in anactual building front, and they have anunobstructed southerly orientation. Theexposure site is kept well mowed. A par-tial view of the three segments of theexposure fence is shown in Figure 5.

Panel inspectionAll panels are inspected every 6

months. To date, we have used onlythree inspectors to minimize inspectionvariability. When installed in October1996, a close-up, black and white photo-graph was taken of each finished panelto photo record its original condition.

An array of test standards has been de-veloped for conducting and evaluatingpaint performance in exterior exposuretest. Pertinent among them are: ASTMD 3274 (ASTM 1982), ASTM D 661(ASTM 1986a), ASTM D 772 (ASTM1986b), ASTM D 1006 (ASTM 1993),and Pictorial Standards of Coatings De-fects (FSCT 1979).

The exposure panel test specimens inthis research are different from thoseused in other research on the weatheringof finished wood. There has been andcontinues to be much public and propri-etary research in weathering perfor-mance of paint specimens. Such re-search involves paint fences wheresmall, rectangular, essentially two-di-

54 APRIL 2003

Table 1. — Specifications for test materials.

Sealant

One-part polyurethane: for construction use to meet Federal Specifications TT-S-00230C and ASTMC 920-94.

Latex designed to meet performance requirements of Federal Specifications TT-S-00230C, Type II,Class A and ASTM C-920-79, Type S, Grade NS, Class 25.

Paint

Latex/latex: primer = water, quartz, acrylic resin, titanium dioxide, alkyd resin, silica cristobalite,silica, isobutyrate ester, zinc oxide, ethylene glycol, octylisothiazolone; top coat = water, acrylic vinylpolymer, titanium dioxide, acrylic resin, propylene glycol, diethylene gylcol, monobutyl ether, silica,octylisothiazolone

Alkyd/latex: primer = talc, mineral spirits, tall oil alkyd polymer, titanium dioxide, soya alkydpolymer, mica, mineral spirits 140-flash, cristobalite, amorphous diatomaceous earth; top coat =water, acrylic polymer, titanium dioxide, zinc oxide, ethylene glycol, quartz, cristobalite

Preservative-treated lumber

Nominal 1-inch southern pine boards: CCA-C 0.40 pcf retention for ground contact

Water-repellent preservatives

Solvent-based and water-based: 3-lodo-2-propynyl butyl carbamate 0.5% and inert ingredients99.5%. Contains petroleum distillates.

Figure 5. — Overall view of exposure site.

mensional uniform specimens of paint-ed wood materials are exposed.

In contrast, our research involvescomplete, three-dimensional construc-tions where paint is but one of five con-struction variables, all of which poten-tially affect the long-term performanceof the painted surface. In addition, thebehavior of joints affects the perfor-mance of the painted surfaces of theexposure panels, just as is the case inactual construction. Therefore, the dev-elopment of a novel inspection and eval-uation procedure is an integral part ofthe research (Wood Research Labora-tory 1996b). Starting with the paintedwood standards just referenced, panelinspection criteria for both surfaces andjoints were developed and are summa-rized in Table 2. To facilitate field in-spection, a template sheet for each de-sign was prepared showing joint andsurface area codes and locations. As anexample, Figure 4 shows the templatefor Design 1. Each panel has it own fieldinspection sheet for data entry. A samplefield inspection sheet for a Design 1exposure panel with associated codingfor surface areas and joints is shown inFigure 6.

Although it varies for the 3 panel de-signs, there are about 480 observationsin the inspection of a single panel, and atotal of about 132,000 observations forthe complete inspection of all panels,which is done every 6 months. For eachlocation and for a given performancecharacteristic, the inspector rates perfor-mance as “no defect” (coded as a 10),“slight to moderate defect” (coded as a6), and “failed condition” (coded as a 2).Clearly, this is a simplification of theusual 10 to 2 rating system of the ASTMtests (ASTM 1986a, 1986b, 1993). Notethat for mildew, the ASTM rating sys-

tem is 10 to 0 (ASTM 1982). However,because of the very large numbers of testpanels and observation sites, it was nec-essary to simplify the rating system. Inthis report, transitions from “no defect”to any other condition (slight to moder-ate defect, or failed condition) are fo-cused upon. In subsequent work, transi-tions from slight to moderate defect tofailed condition will also be analyzed.

Eight performance measures for thepanel surfaces and three performancemeasures for the joints were collectedon six occasions spanning 3 years.These are described in Table 2.

Spreadsheet management of the largeand periodically increasing volume ofinspection data and its preparation forsubsequent statistical analysis are de-tailed in a previous report (Evans 1999).

Data analysisEach performance measure was ana-

lyzed separately using analysis of vari-ance (ANOVA) methods (Neter et al.1996). The ANOVA model includesterms for each of the factors and vari-ables indicating the panel from which anobservation obtains. The latter enablesthe correlation between measurementsmade on the same panel to be taken intoaccount; intuitively this means that eachpanel contributes one observation to in-ferences on factors that are fixed withina panel, and multiple observations (cor-responding to the locations within apanel) to inferences made on factorswhose levels vary within a panel. Thethickness of paint applied was used as acovariate in all analyses, and a locationfactor that identified homogeneous re-gions within a panel was included inmodels for analyses comparing westernredcedar to yellow-poplar. The outcomevariable for the analysis of all factors,

apart from wood type, is the percentageof locations on a given side of a panelthat have yet to degrade to an inferiorquality. For instance, for the perfor-mance characteristic paint cracking, theoutcome is the percentage of locationsthat are rated as having no defect. Theanalysis attempts to attribute variation(across the sides of each panel and be-tween panels) in the percentage of loca-tions that are still in the no defect state,to the factors that vary between the sidesof the panel (i.e., paint color) and be-tween panels (design, surface pretreat-ment, sealant, panel wood, and painttype). Note that the percentage of panelsin the no defect state is the proportion oflocations within a side of a panel thathave not degraded, multiplied by 100.

The ANOVA framework allows inves-tigation of synergies or interactions be-tween factors. Preliminary analysis indi-cated that there were no importantinteractions. Diagnostics, including acareful examination of residuals (Neteret al. 1996), led to the conclusion thatour analytic strategy is appropriate.

ResultsAfter 36 months of weathering, some

differences due to the factors are evi-dent. The major findings are summa-rized in this section. The 0.05 level wasused as the criterion for determining sta-tistical significance. Only comparisons(between levels of the factors) for whichstatistically significant differences werefound are discussed here. For example,the ANOVA model indicated that woodtype was a significant factor for sub-strate cracking. At 36 months, the aver-age percent of locations unchanged withrespect to substrate cracking was 87 forpanels made from yellow-poplar andwestern redcedar versus 56 percent forpanels constructed from dry and wetCCA. A Student Newman Keuls test(see Appendix) indicated that the differ-ence was statistically significant, andthat the difference in outcomes betweenthese groups is the major contributor tothe significant wood type factor.

PretreatmentThe percent changes over time are

plotted in Figure 7. No differences areevident between the water-based and thesolvent-based water-repellent preserva-tive pretreatments. These are both con-sistently superior to the panels that didnot receive any pretreatment. At 36

FOREST PRODUCTS JOURNAL Vol. 53, No. 4 55

Table 2. — Panel inspection criteria.

Surface criteria

Substrate cracking: failure of the wood substrate

Warp: a variation from true or plane surface of any wood component

Rot: the decompositio3n of wood substance caused by fungi

Loose knot: a knot that is no longer held firmly in place

Paint cracking: breaks in the paint film (ASTM D661-86)

Discoloration: the uniform or spotty darkening of the protective or decorative coating

Flaking: detachment of film fragments (ASTM D 772-86)

Mildew: fungus growth that can cause discoloration and ultimate decomposition (ASTM D 3274-82)

Joint criteria

Joint separation: the movement of adjoining parts away from each other

Sealant separation: the movement of the sealant and substrate away from each other

Paint failure (joint): a failure and resultant crack in the paint that covers the sealant in the joint

months, relative to the panels with nopretreatment, the treated panels have su-perior performance for substrate crack-ing (79% vs. 73%), warp (85% vs.76%), paint cracking (60% vs. 52%),paint flaking (95% vs. 91%), joint sepa-ration (48% vs. 30%), and sealantseparation (61% vs. 46%). Althoughpretreatment improved the relative resis-tance to mildew (29% vs. 22%), it is ap-parent from the low values that mildewinfection was overwhelming. The earlyonset of mildew infestation is not shownin Figure 7 but is shown in Figures 8and 10.

Wood

Figure 8 summarizes the results forcomparing four types of wood (yel-low-poplar, western redcedar, dry CCA,and wet CCA) used in the constructionof the panels. Western redcedar was ei-ther the best, or nearly so, of all woodtypes for all surface and joint effects. Asexplained in the above example, for sub-strate cracking, yellow-poplar and west-ern redcedar retained 87 percent versus56 percent for the combined dry and wet

CCA. Western redcedar at 82 percentwas superior for warp resistance versusdry and wet CCA (74%) and yel-low-poplar (67%). For mildew resis-tance, western redcedar retaining 44percent was clearly better than the nextbest wood, which was yellow-poplar at34 percent. For joint separation, westernredcedar at 44 percent was best, fol-lowed by dry CCA (26%), yellow-pop-lar (18%) and wet CCA (10%). Al-though yellow-poplar was the mostdimensionally unstable, as evidenced byits relative warp and joint separationperformance, it surprisingly performedthe best for paint cracking and paintflaking. For paint cracking, the respec-tive retention percentages were yel-low-poplar (68%), western redcedar(62%), and dry and wet CCA (39%).Similarly, for paint flaking, the retentionorder was yellow-poplar (98%), fol-lowed by western redcedar, dry CCA,and wet CCA (all with 93%).

Paint color

The comparison of white with graypaint is shown in Figure 9. White paint

was generally superior to gray paint withrespect to substrate cracking (82% vs.75%), warp (87% vs. 82%), paint crack-ing (63% vs. 53%), and paint flaking(97% vs. 92%).

Paint typeThe all latex and alkyd/latex paint

types are compared in Figure 10. Nei-ther is uniformly superior, with all latexbeing better for substrate cracking (80%vs. 76%), paint cracking (64% vs. 54%),paint flaking (98% vs. 91%), and paintfailure over the sealant (70% vs. 11%).The alkyd/latex type, however, is moreresistant to mildew (35% vs. 21%).

SealantThese data are not shown, but less

warping was observed for those panelsfabricated with latex sealant (88%) thanfor those made with either one-partpolyurethane sealant or no sealant(82%). The purpose of a sealant is tokeep water out of the joints, and sealantseparation measures its effectiveness inthis regard. Latex sealant had less sepa-ration (64%) than the one-part polyure-

56 APRIL 2003

Figure 6. — Field data entry sheet.

thane (48%). Thus the latex sealant’s in-creased effectiveness in excluding waterfrom joints might be a partial explana-tion as to why panels made with it showless warp.

Panel designPanels of Design 2 had less joint sepa-

ration, sealant separation, and sealantpaint failure than the other two designs.In addition, the surface parts of Design 2panels had less degrade due to warp,substrate cracking, and paint cracking.This is of interest considering that De-sign 2 production cost is as low or lowerthan comparable costs for the other twodesigns.

DiscussionThe overall performance of preserva-

tive-treated southern pine was largelyexpected. The basis for inclusion of dryand wet CCA- treated southern pine was

the hypothesis that its long-term decayresistance would overshadow the rela-tively inferior short-term performanceand prevent severe damage, such as thatshown in Figure 1. Since no rot has beenobserved on any of the 274 panels, thishypothesis is still being tested.

Yellow-poplar was dimensionally un-stable, as shown by its tendency to warp.However, the surface of yellow-poplarthat is painted is paradoxically stable, asevidenced by its top-rated paint crackingand paint flaking performance relativeto the other lumber materials. In fact, itequaled the performance of westernredcedar, which is often selected for itsgood paint performance.

Mildew is by far the most active de-grading agent in this weathering test.The incidence of mildew literally ex-ploded after 18 to 24 months of weather-ing. This was an unpleasant surprise as

both commercial paints used in theexperiment were advertised as “mildewresistant.”

A further disappointment with respectto mildew was that the use of either wa-ter-repellent preservative did not practi-cally improve resistance to mildew ascompared to untreated specimens. Thewater-repellent preservatives are adver-tised as providing mildew protection. Inthis case, this benefit was not realized.

Mildew appears as unsightly splot-ches on the paint surface. In addition,with 5× hand lens magnification, it ap-pears that mildew is creating fine fis-sures in the paint surface at some loca-tions. If this initial observation proves tobe true, then the integrity of the paintfilm will be greatly compromised andacceptable performance shortened.

Performance characteristics of thewestern pine trim and MDO panel mate-rials are not reported at this time. West-ern pine trim used on Designs 1 and 2 isconvex in cross section and hence pres-ents a different angle of inclination tosolar rays than a vertical surface. Hencethese materials have a more severeweathering exposure. In subsequentanalysis, these materials and their loca-tions on the face of the exposure panelwill be accounted for and their relativeperformance reported.

The effect of wood types on amountof warp was discussed previously. Thedifference in warp due to selection ofwood types is real, as all panels werefabricated identically. However, from apractical viewpoint, any warp regardlessof wood type used can be greatly less-ened if not eliminated. A light gauge,finish type, nail was used in the auto-nailing gun to attach facing members tothe body of the test panel. These nailswere incapable of holding members inthe plane of the panel under the weather-ing conditions encountered. Subse-quently, identical panels in all respectsexcept choice of nail were built. A de-formed shank nail with a head was usedin fabricating 22 panels that included alltest wood types. These panels have beendisplayed outdoors for over 2-1/2 yearswith no evidence of warp.

After 3 years of weathering, the fol-lowing tentative recommendations areoffered:

• Use a paintable water-repellent pre-servative pretreatment on all un-

FOREST PRODUCTS JOURNAL Vol. 53, No. 4 57

Figure 7. — Weathering performance characteristics vs. surface pretreatment.

painted surfaces and ends of woodparts.

• Use a quality paint system consistingof an acrylic latex primer and twoacrylic latex top coats.

• Use white (or lightly colored) paint, asit is less susceptible to early failurethan dark paints like the gray used inthe study.

• Mildew may be a major problem; adda mildewcide supplement to paint thatis recommended by the paint manu-facturer.

• Use ring-shanked stainless steel nails.

Literature citedAmerican Society for Testing and Materials

(ASTM). 1982 Evaluating degree of surfacedisfigurement of paint films by fungal growthor soil and dirt accumulation. ASTM D3274-82. ASTM, West Conshohocken, PA.

__________. 1986a. Evaluating degree ofcracking of exterior paint. ASTM D 661-86.ASTM, West Conshohocken, PA

__________. 1986b. Evaluating degree of flak-ing (scaling) of exterior paints. ASTM D772-86. ASTM, West Conshohocken, PA.

__________. 1993. Standard practice for con-ducting exterior exposure tests of paints onwood. ASTM D 1006-93. ASTM, WestConshohocken, PA.

Christensen, R.R. 1996. Plane Answers toComplex Questions. Springer, New York.

City of Lafayette (Indiana). The Main Streetprogram for facade renovation of downtownbuildings. Community Development and Re-development Dept. www.city.lafay-ette.in.us/government/departments/cdrd/mainstreet.htm.

Evans, J.W. 1999. Data management and analy-sis handbook for historic restoration researchproject, Wood Research Laboratory PurdueUniversity, West Lafayette, IN.www.fnr.purdue.edu/woodresearch/facade/in-dex.html.

Federation of Societies for Coatings Technol-ogy (FSCT). 1979. Pictorial standards forcoatings defects. FSCT, Blue Bell, PA

Feist, W.C. 1990. Weathering performance ofpainted wood pretreated with water-repellentpreservatives. Forest Prod. J. 40(7/8) 21-26.

Neter, J., M.H. Kuter, C.J. Nachtsheim, and W.Wasserman. 1996. Applied Linear StatisticalModels. Irwin, Chicago, IL.

Williams, R.S., M.T. Knaebe, and W.C. Feist.1996. Finishes for exterior wood: Selection,application, and maintenance. Pub. No. 7291.Forest Prod. Soc., Madison, WI. 127 pp.

Wood Research Laboratory. 1996a. Wood in his-toric preservation and renovation: Panel con-struction specifications. Purdue Univ., West La-fayette, IN. www.fnr.purdue.edu/woodresearch/facade/index.html

__________. 1996b. Panel performance evalu-ation. Purdue Univ., West Lafayette, IN.www.fnr.purdue.edu/woodresearch/fa-cade/index.html

58 APRIL 2003

Figure 9. — Weathering performance characteristics vs. paint color.

Figure 8. — Weathering performance characteristics vs. wood.

AppendixAnalysis of variance (ANOVA) tables and

associated statistics are presented for jointseparation and warp performance measures.Output such as this was used to determinewhich factors had significant bearing on theperformance of the panels, and to evaluatestatistical tests between levels or groups oflevels of these factors.

Joint separationJoint separation is a performance measure

evaluated at joints on the panel. A complica-tion arises when two different types of woodmeet at a joint. This was overcome by re-stricting the analysis to just those joints forwhich the wood types are the same. Thisdrastically reduced the number of observa-tions. The number of observations was fur-

ther reduced by considering only four woodtypes: western redcedar, yellow-poplar, dryCCA, and wet CCA; western pine (used forthe trim) and MDO (used as the base panel)were excluded. A total of 1,138 locationsacross all the panels had valid observationsfor joint separation at 36 months. Thus, a to-tal of 1,137 degrees of freedom were avail-able for model fitting. Because multiple ob-servations are taken within each panel, theclustering of observations within a panelmust be modeled.

The Location factor identifies homoge-neous regions within each side of the panel.Because the locations vary with the design ofthe panel, Location is nested in Design (i.e.,distinct parameters are associated with eachdesign). As mentioned in the body of the ar-ticle, Wood is a within-panel effect becausethe type of wood used varies within a panel.

Care must be taken to separate panel-leveleffects from within-panel effects. Factorssuch as design, surface pretreatment, sealant,and paint type are constant within a paneland so must be compared between panels.The correct F-test for these effects is basedon the marginal variance between panels, es-sentially a sum of the random variation at-tributed to panel and pure error variation.Because the design of the experiment is notbalanced with respect to the factors, the de-nominator depends on the effect being tested(in the numerator of the F-test), but is ap-proximately equal to the Error (betweenpanel) term. Factors that vary within panelsare evaluated using the usual F-test; i.e., us-ing the pure error variance in the denomina-tor. A consequence of the unbalanced designis that a portion of the variation in the out-comes cannot be uniquely partitioned amongthe effects. The results presented herein areconservative in that only the variationuniquely attributable to each factor is used toestablish statistical significance. The analy-sis of joint separation is based on Table A1.

In the analysis of joint separation, Sealant,Paint type, Paint color, and Wood are signifi-cant at the 0.01 level (indicating very strongeffects). The control variables Side andThickness of paint are also significant (theinclusion of these variables reduces the errorsum of squares, resulting in more powerfulwithin-panel significance tests).

For the purpose of illustration, particularattention is paid to the effect of Wood. With asignificant effect having been found, the lev-els of Wood may be compared to determinewhere the effect occurs. This is performedusing the Student-Newman-Keuls (SNK)test (Christensen 1996). This is a multiplecomparison test that determines groups ofthe levels of a factor between which statisti-cally significant differences preside.

The results of the SNK test for wood forthe joint separation performance measureare summarized in Table A2.

Table A2 shows the groupings of woodfor which significant differences are foundbetween the resulting groups. Each level ofwood is grouped by itself, indicating that thewood types perform differently from eachother. Western redcedar performs the best,followed by Dry CCA, Yellow-poplar, andWet CCA.

FOREST PRODUCTS JOURNAL Vol. 53, No. 4 59

Table A1. — Analysis of joint separation.a

Source DF SS MS F-value p-value

Design 2 3081 1540.4 1.43 0.4497

Surface pretreatment 2 3320 1659.9 1.54 0.3886

Sealant 2 120369 60184.5 55.8 0.0001

Paint type 1 12042 12041.8 11.2 0.0059

Error (between panel) 264 540673 2048 1.90 0.0001

Paint color 1 19076 19076.2 17.7 0.0001

Side 1 7777 7776.7 7.21 0.0043

Location (design) 5 8375 1675.1 1.55 0.1769

Wood 1 46750 46749.7 43.3 0.0001

Thickness of paint 1 5059 5058.9 4.69 0.0306

Error (within panel) 855 922524 1079

Total 1137 1928225aDF = degrees of freedom; SS = sum of squares; MS = mean square.

Figure 10. — Weathering performance characteristics vs. paint type.

Table A2. — Results of the Student-Newman-Keuls test for wood.

Group Group mean No. obs. Wood

A 47.2 123 Western redcedar

B 26.0 480 Dry CCA

C 18.7 123 Yellow-poplar

D 9.7 412 Wet CCA

Warp (surfaceperformance measure)

In the example presented here, the factorof main interest is Surface pretreatment.Therefore, we can simplify the analysis byusing the panel wood factor (constant with apanel), as opposed to the wood factor, andanalyze the proportion of locations of a sideof a panel that have degraded. This meansthat Location can be removed from themodel, leaving Side as the only factor re-quired to account for the whereabouts of themeasurement. The associated ANOVA isshown in Table A3.

Design, Surface pretreatment, Sealant,and Paint color are the significant factors inthe performance of the panels with respect towarp. The level of significance in each caseexceeds 0.01, indicating that the effects arestrong. The effect of surface pretreatment isinvestigated in more detail using the SNKtest (Table A4).

The SNK test indicates that there is a sig-nificant difference (at the 0.05 level) be-tween pretreatment and no pretreatment.However, there is not sufficient evidence toconclude that there is a difference in perfor-mance between the VOC and Classicpretreatments.

60 APRIL 2003

Table A4. — Results of the Student-Newman-Keuls test for warp.

Group Group mean No. obs. Surface pretreatment

A 86.2 242 VOC

A 84.2 246 Classic

B 76.3 60 None

Table A3. — ANOVA results for warp.a

Source DF SS MS F-value p-value

Design 2 1173 586.4 12.5 0.0087

Surface pretreatment 2 1793 896.4 19.1 0.0005

Sealant 2 1814 906.9 19.23 0.0003

Panel wood 2 177 88.5 1.88 0.4269

Paint type 1 49.3 49.3 1.05 0.4685

Error (between panel) 264 33566 127.15 2.70 0.0001

Paint color 1 2183 2183.4 46.4 0.0001

Side 1 28.6 28.6 0.61 0.436

Thickness of paint 1 0 0 0 0.9989

Error (within panel) 271 12743 47.0

Total 547 59382aDF = degrees of freedom; SS = sum of squares; MS = mean square.