Author: andrew

Fall 2018 Courses

Teaching, by andrew, August 31, 2018

Organizational Field Research
Doctoral Course

The purpose of this course is to immerse you in the discipline and practice of research on organizations “in the wild.” We will do this through readings, class discussions, and—most importantly—your experiences embedded within an organization throughout the semester. You will use reflective memos and feedback from faculty and your classmates to develop an initial mental model of what it means to develop and advance research projects that are grounded both in theory and in how real organizations operate today. The key learning objectives of the course are:

  • To build your mental model of what it means to be an organizational researcher.
  • To lay the necessary foundations for conducting ethical research in organizations.
  • To develop your skills in observing organizations, integrating your observations with existing theory, and developing relevant research projects.
  • To enhance your understanding of what constitutes an academic contribution.

Leadership Development
2nd Year Full-Time MBA OB Core Course

This course builds upon the material from the 1st Year OB Core (OB 5620, Foundations for Leadership Effectiveness) and, importantly, from your time so far at Olin and during your summer work experiences. The focus of the course is on the attributes, behaviors, and tendencies of effective leadership. There are two primary objectives:

  • Gain new insights into your own beliefs and expectations regarding what constitutes effective leadership in organizations. You will accomplish this through a mixture of classroom discussion, case analysis, and self-assessment.
  • Learn about your own strengths and weaknesses in leading others. You will accomplish this in the classroom through controlled experiential exercises, which will be the basis for feedback from your peers. You will also reflect in depth on your strengths using feedback provided by people you have encountered in your life and career through a structured exercise.

Foundations for Leadership Effectiveness
1st Year Full-Time MBA OB Core Course

This course presents a framework for thinking about how individual attributes and interpersonal skills serve as a foundation for effective leadership in small groups and teams. Through experiential exercises and classroom discussions, this course will enable you to gain deeper insights into your current strengths as a leader and to identify developmental opportunities for the future. There are two primary objectives:

  • Deepen your self-awareness by enhancing your insights into (1) your personal characteristics and attributes; (2) your interpersonal, social, and leadership skills; and, (3) your approach in working within groups and teams.
  • Improve your leadership effectiveness by enhancing your capacity to (1) identify your own leadership strengths and weaknesses and (2) understand how your assets and liabilities combine with others’ assets and liabilities in team-based work.

Response Surface Analysis with Clustered Standard Errors

Code, by andrew, January 8, 2018

I wanted to easily use response surface analysis, but with clustered standard errors and also with the possibility of adding control variables to the model. So, I hacked the RSA package in R and expanded its functionality a bit. A few notes to keep in mind when using this:

  • These changes will only apply for models = c(“full”). None of the other models are supported right now.
  • The control variable functionality was already in the package, but was disabled by the authors because it’s not propagated throughout all models. As with clustered standard errors, I have only enabled this within the “full” model.
  • The output will include an option in the models list called fullcluster. Access the object as follows: out$models$fullcluster
  • I will add more detail when I have some time.
## ---------------------------- ##
## This is a modified version of the RSA function. The purpose here
## is to provide robust clustered robust standard errors
## and to use control variables for at least the full model.
## ---------------------------- ##

RSA.akmod <- function (formula, data = NULL, center = FALSE, scale = FALSE,
    na.rm = FALSE, out.rm = TRUE, breakline = FALSE, models = "default",
    cubic = FALSE, verbose = TRUE, add = "", estimator = "MLR",
    se = "robust", missing = NA, ..., control.variables = c(), cluster.variable = c())
{
    require(RSA)
## ---------------------------- ##
## The original version of the function excludes control variable functionality
## I am going to implement it at least for the full model for now.
## ---------------------------- ##
#    if (length(control.variables) > 0)
#        stop("Control.variables feature not implemented yet!")
## ---------------------------- ##

## Editing this to include the fullcluster model ##

    validmodels <- c("absdiff", "absunc", "diff", "mean", "additive",
        "IA", "SQD", "SRRR", "SRR", "RR", "SSQD", "SRSQD", "full",
        "null", "onlyx", "onlyy", "onlyx2", "onlyy2", "weak",
        "strong", "fullcluster")
    if (length(models) == 1 & models[1] == "all") {
        models <- validmodels
    }
    if (length(models) == 1 & models[1] == "default") {
        models <- c("additive", "IA", "SQD", "SRRR", "SRR", "RR",
            "SSQD", "SRSQD", "full", "null", "onlyx2", "onlyy2",
            "onlyx", "onlyy")
    }
    if (any(!models %in% validmodels))
        stop("Unknown model name provided in parameter 'models'.")
    s.NULL <- s.full <- s.full.cluster <- s.IA <- s.diff <- s.mean <- s.absdiff <- s.additive <- s.SQD <- s.SSQD <- s.SRSQD <- s.absunc <- s.cubic <- s.RR <- s.SRR <- s.SRRR <- s.onlyx <- s.onlyy <- s.onlyx2 <- s.onlyy2 <- s.weak <- s.strong <- NULL
    SRSQD.rot <- ""
    SRRR.rot <- ""
    add <- paste0("\n# User defined syntax:\n", add)
   
## ---------------------------- ##
## This section of RSA creates scaled variables, creates the polynomial terms, checks the range of variables
## and checks for missing values
## ---------------------------- ##    
    DV <- all.vars(formula)[1]
    IV1 <- all.vars(formula)[2]
    IV2 <- all.vars(formula)[3]
    df <- data[, c(DV, IV1, IV2, control.variables, cluster.variable)]
    df[, IV1] <- scale(df[, IV1], center = center, scale = scale)
    df[, IV2] <- scale(df[, IV2], center = center, scale = scale)
    df <- add.variables(formula, data.frame(data.matrix(df)))
    if (0 < min(df[, IV1], na.rm = TRUE) | 0 > max(df[, IV1],
        na.rm = TRUE))
        warning(paste("The numerical zero point is outside of the range of variable",
            IV1, ". Please consider re-centering the variable."))
    if (0 < min(df[, IV2], na.rm = TRUE) | 0 > max(df[, IV2],
        na.rm = TRUE))
        warning(paste("The numerical zero point is outside of the range of variable",
            IV2, ". Please consider re-centering the variable."))
    if ((max(df[, IV1], na.rm = TRUE) - min(df[, IV1], na.rm = TRUE))/(max(df[,
        IV2], na.rm = TRUE) - min(df[, IV2], na.rm = TRUE)) >
        2)
        warning("Predictor variables have a very different range (by factor 2 or larger)- please check scaling of variables.")
    if (is.na(missing)) {
        if (any(is.na(df))) {
            missing <- "fiml"
            warning("There are missing values in your data set. Model is computed with option `missing = 'fiml'`. This is only valid if the data are missing completely at random (MCAR) or missing at random (MAR)! If you want to exclude NA, use `missing = 'listwise'`",
                call. = FALSE)
        }
        else {
            missing <- "listwise"
        }
    }
## ---------------------------- ##
## This section of RSA creates the string names
## of the newly created variables (above) for higher order terms and
## interaction terms. This also creates the addition for control variables.
## ---------------------------- ##    
    IV12 <- paste0(IV1, "2")
    IV22 <- paste0(IV2, "2")
    IV13 <- paste0(IV1, "3")
    IV23 <- paste0(IV2, "3")
    IV_IA <- paste0(IV1, "_", IV2)
    IV_IA2 <- paste0(IV1, "_", IV2, "2")
    IV_IA3 <- paste0(IV1, "2", "_", IV2)
    W_IV1 <- paste0("W_", IV1)
    W_IV2 <- paste0("W_", IV2)
    CV <- ifelse(length(control.variables > 0), paste0(" + ",
        paste(control.variables, collapse = " + ")), "")
    addcubic <- ""
    if (cubic == TRUE)
        addcubic <- paste0(" + ", paste(IV13, IV23, IV_IA2, IV_IA3,
            sep = " + "))
    f <- paste0(paste0(DV, " ~ ", paste(IV1, IV2, IV12, IV_IA,
        IV22, sep = " + ")), addcubic, CV)
       
## ---------------------------- ##
# This uses regression to get model statistics and examine for outliers
## ---------------------------- ##        

## ---------------------------- ##        
# AK NOTE: Need to modify this to provide the summary statistics for the model
# that has the control variables included. Maybe include an additional model
# that way we can have a change in the F r2 test from control to
# inclusion of the polynomial terms.
## ---------------------------- ##        

    lm.full <- lm(f, df, na.action = na.exclude)
    if (is.null(out.rm) || (typeof(out.rm) == "logical" && out.rm ==
        TRUE)) {
        out.rm <- "bj1980"
    }
    if ((typeof(out.rm) == "logical" && out.rm == FALSE)) {
        out.rm <- "none"
    }
    out.rm <- match.arg(out.rm, c("bj1980", "robust", "none"))
    df$out <- FALSE
    if (out.rm == "bj1980") {
        inf <- influence.measures(lm.full)
        df$out <- apply(inf$is.inf[, c("dffit", "cook.d", "hat")],
            1, sum) == 3
        n.out <- sum(na.omit(df$out) == TRUE)
        if (verbose == TRUE & n.out > 0) {
            warning(paste("Removed", n.out, "multivariate outlier(s) according to Bollen & Jackman (1980) criteria. Outliers are in row(s):",
                paste(which(df$out == TRUE), collapse = ", ")))
        }
    }
    if (out.rm == "robust") {
        stop("Robust outlier detection not implemented yet.")
    }
    df$out[is.na(df$out)] <- FALSE
## ---------------------------- ##
# This section of RSA builds the polynomial equations and runs the
# path analysis.
## ---------------------------- ##    
   
    withCallingHandlers({
        poly <- paste0(DV, " ~ b1*", IV1, " + b2*", IV2, " + b3*",
            IV12, " + b4*", IV_IA, " + b5*", IV22, CV)
        if ("null" %in% models) {
            s.NULL <- sem(paste0(DV, "~ 1 + 0*", IV1, " + 0*",
                IV2, " + 0*", IV12, " + 0*", IV_IA, " + 0*",
                IV22, CV), data = df[df$out == FALSE, ], fixed.x = TRUE,
                meanstructure = TRUE, se = se, estimator = estimator,
                missing = missing, ...)
        }
        if ("additive" %in% models) {
            if (verbose == TRUE)
                print("Computing additive model (additive) ...")
            m.additive <- paste(poly, "b3==0", "b4==0", "b5==0",
                "a1 := b1+b2", "a2 := b3+b4+b5", "a3 := b1-b2",
                "a4 := b3-b4+b5", "a5 := b3-b5", add, sep = "\n")
            s.additive <- sem(m.additive, data = df[df$out ==
                FALSE, ], fixed.x = TRUE, meanstructure = TRUE,
                se = se, estimator = estimator, missing = missing,
                ...)
        }
        if ("onlyx2" %in% models) {
            if (verbose == TRUE)
                print("Computing x + x^2 model (onlyx2) ...")
            m.onlyx2 <- paste(poly, "b2==0", "b4==0", "b5==0",
                "a1 := b1+b2", "a2 := b3+b4+b5", "a3 := b1-b2",
                "a4 := b3-b4+b5", "a5 := b3-b5", add, sep = "\n")
            s.onlyx2 <- sem(m.onlyx2, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("onlyy2" %in% models) {
            if (verbose == TRUE)
                print("Computing y + y^2 model (onlyy2) ...")
            m.onlyy2 <- paste(poly, "b1==0", "b3==0", "b4==0",
                "a1 := b1+b2", "a2 := b3+b4+b5", "a3 := b1-b2",
                "a4 := b3-b4+b5", "a5 := b3-b5", add, sep = "\n")
            s.onlyy2 <- sem(m.onlyy2, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("onlyx" %in% models) {
            if (verbose == TRUE)
                print("Computing x model (onlyx) ...")
            m.onlyx <- paste(poly, "b2==0", "b3==0", "b4==0",
                "b5==0", "a1 := b1+b2", "a2 := b3+b4+b5", "a3 := b1-b2",
                "a4 := b3-b4+b5", "a5 := b3-b5", add, sep = "\n")
            s.onlyx <- sem(m.onlyx, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("onlyy" %in% models) {
            if (verbose == TRUE)
                print("Computing y model (onlyy) ...")
            m.onlyy <- paste(poly, "b1==0", "b3==0", "b4==0",
                "b5==0", "a1 := b1+b2", "a2 := b3+b4+b5", "a3 := b1-b2",
                "a4 := b3-b4+b5", "a5 := b3-b5", add, sep = "\n")
            s.onlyy <- sem(m.onlyy, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("diff" %in% models) {
            if (verbose == TRUE)
                print("Computing difference model (diff) ...")
            m.diff <- paste(poly, "b3==0", "b4==0", "b5==0",
                "b1 == -b2", "a1 := b1+b2", "a2 := 0", "a3 := b1-b2",
                "a4 := 0", add, sep = "\n")
            s.diff <- sem(m.diff, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("mean" %in% models) {
            if (verbose == TRUE)
                print("Computing mean model (mean) ...")
            m.mean <- paste(poly, "b3==0", "b4==0", "b5==0",
                "b1 == b2", "a1 := b1+b2", "a2 := 0", "a3 := b1-b2",
                "a4 := 0", add, sep = "\n")
            s.mean <- sem(m.mean, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("IA" %in% models) {
            if (verbose == TRUE)
                print("Computing interaction model (IA)...")
            m.IA <- paste(poly, "b3==0", "b5==0", "a1 := b1+b2",
                "a2 := b3+b4+b5", "a3 := b1-b2", "a4 := b3-b4+b5",
                "a5 := b3-b5", "X0 := (b2*b4 - 2*b1*b5) / (4*b3*b5 - b4^2)",
                "Y0 := (b1*b4 - 2*b2*b3) / (4*b3*b5 - b4^2)",
                "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "p10 := Y0 - p11*X0", "p20 := Y0 - p21*X0", "PA1.curv := b3 + b4*p11 + b5*(p11^2)",
                "PA2.curv := b3 + b4*p21 + b5*(p21^2)", "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.IA <- sem(m.IA, data = df[df$out == FALSE, ], fixed.x = TRUE,
                meanstructure = TRUE, se = se, estimator = estimator,
                missing = missing, ...)
        }
        if ("SQD" %in% models) {
            if (verbose == TRUE)
                print("Computing squared difference model (SQD) ...")
            m.SQD <- paste(poly, "b1==0", "b2==0", "b3==b5",
                "b3+b4+b5==0", "a1 := b1+b2", "a2 := b3+b4+b5",
                "a3 := b1-b2", "a4 := b3-b4+b5", "a5 := b3-b5",
                "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "PA1.curv := b3 + b4*p11 + b5*(p11^2)", "PA2.curv := b3 + b4*p21 + b5*(p21^2)",
                "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.SQD <- sem(m.SQD, data = df[df$out == FALSE, ],
                fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("SSQD" %in% models) {
            if (verbose == TRUE)
                print("Computing shifted squared difference model (SSQD) ...")
            m.SSQD <- paste(poly, "b1==-b2", "b3==b5", "b3+b4+b5==0",
                "a1 := b1+b2", "a2 := b3+b4+b5", "a3 := b1-b2",
                "a4 := b3-b4+b5", "a5 := b3-b5", "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "PA1.curv := b3 + b4*p11 + b5*(p11^2)", "PA2.curv := b3 + b4*p21 + b5*(p21^2)",
                "C := b1 / (2*b3)", "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.SSQD <- sem(m.SSQD, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if (any(models %in% c("RR"))) {
            if (verbose == TRUE)
                print("Computing rising ridge model (RR) ...")
            m.RR <- paste(poly, "b1==b2", "b3==b5", "b3+b4+b5==0",
                "a1 := b1+b2", "a2 := b3+b4+b5", "a3 := b1-b2",
                "a4 := b3-b4+b5", "a5 := b3-b5", "meaneffect := b1+b2",
                "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "PA1.curv := b3 + b4*p11 + b5*(p11^2)", "PA2.curv := b3 + b4*p21 + b5*(p21^2)",
                "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.RR <- sem(m.RR, data = df[df$out == FALSE, ], fixed.x = TRUE,
                meanstructure = TRUE, se = se, estimator = estimator,
                missing = missing, ...)
        }
        if (any(models %in% c("SRR"))) {
            if (verbose == TRUE)
                print("Computing shifted rising ridge model (SRR) ...")
            m.SRR <- paste(poly, "b3==b5", "b3+b4+b5==0", "a1 := b1+b2",
                "a2 := b3+b4+b5", "a3 := b1-b2", "a4 := b3-b4+b5",
                "a5 := b3-b5", "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "PA1.curv := b3 + b4*p11 + b5*(p11^2)", "PA2.curv := b3 + b4*p21 + b5*(p21^2)",
                "meaneffect := a1", "C := (b1-b2) / (4*b3)",
                "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.SRR <- sem(m.SRR, data = df[df$out == FALSE, ],
                fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if (any(models %in% c("SRRR"))) {
            if (verbose == TRUE)
                print("Computing rotated and shifted rising ridge model (SRRR), up ...")
            m.SRRR.up <- paste(paste(poly, " + start(0.01)*",
                IV12, " + start(0.01)*", IV22), "b3 > 0.000001",
                "b5 > 0.000001", "b4^2 == 4*b3*b5", "a1 := b1+b2",
                "a2 := b3+b4+b5", "a3 := b1-b2", "a4 := b3-b4+b5",
                "a5 := b3-b5", "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "PA1.curv := b3 + b4*p11 + b5*(p11^2)", "PA2.curv := b3 + b4*p21 + b5*(p21^2)",
                "meaneffect := (b2*b4 - 2*b1*b5) / b4", "C := (-2*b1*b5 - b2*b4) / (4*b4*b5)",
                "S := (-b4) / (2*b5)", "a4.rescaled := b3/S^2 - b4/S + b5",
                "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.SRRR.up <- sem(m.SRRR.up, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
            if (verbose == TRUE)
                print("Computing rotated and shifted rising ridge model (SRRR), down ...")
            m.SRRR.down <- paste(paste(poly, " + start(-0.01)*",
                IV12, " + start(-0.01)*", IV22), "b3 < -0.000001",
                "b5 < -0.000001", "b4^2 == 4*b3*b5", "a1 := b1+b2",
                "a2 := b3+b4+b5", "a3 := b1-b2", "a4 := b3-b4+b5",
                "a5 := b3-b5", "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "PA1.curv := b3 + b4*p11 + b5*(p11^2)", "PA2.curv := b3 + b4*p21 + b5*(p21^2)",
                "meaneffect := (b2*b4 - 2*b1*b5) / b4", "C := (-2*b1*b5 - b2*b4) / (4*b4*b5)",
                "S := (-b4) / (2*b5)", "a4.rescaled := b3/S^2 - b4/S + b5",
                "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.SRRR.down <- sem(m.SRRR.down, data = df[df$out ==
                FALSE, ], fixed.x = TRUE, meanstructure = TRUE,
                se = se, estimator = estimator, missing = missing,
                ...)
            if (inspect(s.SRRR.up, "converged") == FALSE & inspect(s.SRRR.down,
                "converged") == TRUE) {
                SRRR.rot <- "down"
            }
            else if (inspect(s.SRRR.up, "converged") == TRUE &
                inspect(s.SRRR.down, "converged") == FALSE) {
                SRRR.rot <- "up"
            }
            else if (inspect(s.SRRR.up, "converged") == TRUE &
                inspect(s.SRRR.down, "converged") == TRUE) {
                SRRR.rot <- ifelse(fitMeasures(s.SRRR.up, "chisq") >
                  fitMeasures(s.SRRR.down, "chisq"), "down",
                  "up")
            }
            else {
                if (verbose == TRUE)
                  print("Warning: SRRR model has not converged (neither up nor down curvature)")
            }
            if (SRRR.rot == "up") {
                s.SRRR <- s.SRRR.up
            }
            else if (SRRR.rot == "down") {
                s.SRRR <- s.SRRR.down
            }
            if (verbose == TRUE)
                print(paste0("Direction of SRRR curvature: ",
                  SRRR.rot))
        }
        if (any(models %in% c("SRSQD"))) {
            if (verbose == TRUE)
                print("Computing rotated squared difference model (SRSQD), up ...")
            m.SRSQD.up <- paste(paste(poly, " + start(0.001)*",
                IV22), "b1 == (b2*b4)/(2*b5)", "b3 > 0.000001",
                "b5 > 0.000001", "b4^2 == 4*b3*b5", "C := -.5*(b2/b5)",
                "S := (-b4) / (2*b5)", "a1 := b1+b2", "a2 := b3+b4+b5",
                "a3 := b1-b2", "a4 := b3-b4+b5", "a5 := b3-b5",
                "a4.rescaled := b3/S^2 - b4/S + b5", "X0 := (b2*b4 - 2*b1*b5) / (4*b3*b5 - b4^2)",
                "Y0 := (b1*b4 - 2*b2*b3) / (4*b3*b5 - b4^2)",
                "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "p10 := Y0 - p11*X0", "p20 := Y0 - p21*X0", "PA1.curv := b3 + b4*p11 + b5*(p11^2)",
                "PA2.curv := b3 + b4*p21 + b5*(p21^2)", "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.SRSQD.up <- sem(m.SRSQD.up, data = df[df$out ==
                FALSE, ], fixed.x = TRUE, meanstructure = TRUE,
                se = se, estimator = estimator, missing = missing,
                ...)
            if (verbose == TRUE)
                print("Computing rotated squared difference model (SRSQD), down ...")
            m.SRSQD.down <- paste(paste(poly, " + start(-0.001)*",
                IV22), "b1 == (b2*b4)/(2*b5)", "b3 < -0.000001",
                "b5 < -0.000001", "b4^2 == 4*b3*b5", "C := -.5*(b2/b5)",
                "S := (-b4) / (2*b5)", "a1 := b1+b2", "a2 := b3+b4+b5",
                "a3 := b1-b2", "a4 := b3-b4+b5", "a5 := b3-b5",
                "a4.rescaled := b3/S^2 - b4/S + b5", "X0 := (b2*b4 - 2*b1*b5) / (4*b3*b5 - b4^2)",
                "Y0 := (b1*b4 - 2*b2*b3) / (4*b3*b5 - b4^2)",
                "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p10 := Y0 - p11*X0", "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "p20 := Y0 - p21*X0", "PA1.curv := b3 + b4*p11 + b5*(p11^2)",
                "PA2.curv := b3 + b4*p21 + b5*(p21^2)", "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                add, sep = "\n")
            s.SRSQD.down <- sem(m.SRSQD.down, data = df[df$out ==
                FALSE, ], fixed.x = TRUE, meanstructure = TRUE,
                se = se, estimator = estimator, missing = missing,
                ...)
            if (inspect(s.SRSQD.up, "converged") == FALSE & inspect(s.SRSQD.down,
                "converged") == TRUE) {
                SRSQD.rot <- "down"
            }
            else if (inspect(s.SRSQD.up, "converged") == TRUE &
                inspect(s.SRSQD.down, "converged") == FALSE) {
                SRSQD.rot <- "up"
            }
            else if (inspect(s.SRSQD.up, "converged") == TRUE &
                inspect(s.SRSQD.down, "converged") == TRUE) {
                SRSQD.rot <- ifelse(fitMeasures(s.SRSQD.up, "chisq") >
                  fitMeasures(s.SRSQD.down, "chisq"), "down",
                  "up")
            }
            else {
                if (verbose == TRUE)
                  warning("Warning: SRSQD model has not converged (neither up nor down curvature)")
            }
            if (SRSQD.rot == "up") {
                s.SRSQD <- s.SRSQD.up
            }
            else if (SRSQD.rot == "down") {
                s.SRSQD <- s.SRSQD.down
            }
            if (verbose == TRUE)
                print(paste0("Direction of SRSQD curvature: ",
                  SRSQD.rot))
        }
## ---------------------------- ##
## Here is the polynomial model that I'm going to alter. It is going to use
## clustered robust standard errors (if the user specified a clustering variable
## ---------------------------- ##        

        if ("full" %in% models) {
            if (verbose == TRUE)
                print("Computing polynomial model (full) ...")
            m.full <- paste(poly, "a1 := b1+b2", "a2 := b3+b4+b5",
                "a3 := b1-b2", "a4 := b3-b4+b5", "a5 := b3-b5",
                "X0 := (b2*b4 - 2*b1*b5) / (4*b3*b5 - b4^2)",
                "Y0 := (b1*b4 - 2*b2*b3) / (4*b3*b5 - b4^2)",
                "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p10 := Y0 - p11*X0", "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "p20 := Y0 - p21*X0", "PA1.curv := b3 + b4*p11 + b5*(p11^2)",
                "PA2.curv := b3 + b4*p21 + b5*(p21^2)", "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "weakcondition    := b3*b5", "strongcondition1 := (b2*b4)/(2*b5) - b1",
                "strongcondition2 := 2*sqrt(b3*b5)  - b4", add,
                sep = "\n")
               
            # This model is not going to deal with missing values in the way that is specified above. It will just use the default for SEM, which is dependent upon the type of estimator that is used.
           
            # Have to actually create a full string of this so that the full call is included in the s.full object. If I just use m.full in a regular sem call, then I'll get m.full in the output object
           
             call.full = paste("sem(model='",m.full,"', data=df[df$out == FALSE, ], fixed.x=TRUE, meanstructure=TRUE, se='",se,"', estimator='",estimator,"', ...)", sep="")
             
             str_eval <- function(x) {return(eval(parse(text=x)))}
             s.full <- str_eval(call.full)
                       
            ## ------------------ ##
            ## This is the only change, but it creates an additional model
            ## to report
            ## ------------------ ##           
                           
            if("fullcluster" %in% models) {                            
                d2 = svydesign(ids=~get(cluster.variable), data=df)            
                s.full.cluster = lavaan.survey(s.full, survey.design=d2, estimator=estimator)
           
            }                
        }
       
        if ("weak" %in% models) {
            if (verbose == TRUE)
                print("Computing weak fit pattern ...")
            m.weak <- paste(poly, "a1 := b1+b2", "a2 := b3+b4+b5",
                "a3 := b1-b2", "a4 := b3-b4+b5", "a5 := b3-b5",
                "X0 := (b2*b4 - 2*b1*b5) / (4*b3*b5 - b4^2)",
                "Y0 := (b1*b4 - 2*b2*b3) / (4*b3*b5 - b4^2)",
                "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p10 := Y0 - p11*X0", "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "p20 := Y0 - p21*X0", "PA1.curv := b3 + b4*p11 + b5*(p11^2)",
                "PA2.curv := b3 + b4*p21 + b5*(p21^2)", "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "b3*b5 > 0", add, sep = "\n")
            s.weak <- sem(m.weak, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("strong" %in% models) {
            if (verbose == TRUE)
                print("Computing strong fit pattern ...")
            m.strong <- paste(poly, "a1 := b1+b2", "a2 := b3+b4+b5",
                "a3 := b1-b2", "a4 := b3-b4+b5", "a5 := b3-b5",
                "p11 := (b5 - b3 + sqrt(((b3 - b5)^2) + (b4^2))) / b4",
                "p21 :=  (b5 - b3 - sqrt((b3 - b5)^2 + b4^2)) / b4",
                "PA1.curv := b3 + b4*p11 + b5*(p11^2)", "PA2.curv := b3 + b4*p21 + b5*(p21^2)",
                "l1 := (b3 + b5 + sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "l2 := (b3 + b5 - sqrt((b3+b5)^2 - 4*b3*b5 + b4^2))/2",
                "b3*b5 > 0.000001", "(b2*b4) == 2*b1*b5", "4*b3*b5  == b4^2",
                add, sep = "\n")
            s.strong <- sem(m.strong, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if (cubic == TRUE) {
            if (verbose == TRUE)
                print("Computing full cubic model (cubic) ...")
            m.cubic <- paste(paste0(poly, " + b9*", IV13, " + b10*",
                IV_IA2, " + b11*", IV_IA3, " + b12*", IV23),
                "u1 := b1 + b2", "u2 := b3 + b4 + b5", "u3 := b9 + b10 + b11 + b12",
                "v1 := b1 - b2", "v2 := b3 - b4 + b5", "v3 := b9 + b10 - b11 - b12",
                add, sep = "\n")
            s.cubic <- sem(m.cubic, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("absdiff" %in% models) {
            if (verbose == TRUE)
                print("Computing constrained absolute difference model (absdiff) ...")
            m.absdiff <- paste(paste0(DV, " ~ b1*", IV1, " + b2*",
                IV2, " + b6*W + b7*W_", IV1, " + b8*W_", IV2),
                "b1 == 0", "b2 == 0", "b6 == 0", "b7 == -b8",
                add, sep = "\n")
            s.absdiff <- sem(m.absdiff, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
        if ("absunc" %in% models) {
            if (verbose == TRUE)
                print("Computing unconstrained absolute difference model (absunc) ...")
            m.absunc <- paste(paste0(DV, " ~ b1*", IV1, " + b2*",
                IV2, " + b6*W + b7*W_", IV1, " + b8*W_", IV2),
                ifelse(breakline == FALSE, "b6==0", ""), add,
                sep = "\n")
            s.absunc <- sem(m.absunc, data = df[df$out == FALSE,
                ], fixed.x = TRUE, meanstructure = TRUE, se = se,
                estimator = estimator, missing = missing, ...)
        }
    }, warning = function(w) {
        W <- as.character(w$call)
        if ((W[1] == "sqrt" & W[2] == "diag(def.cov)" & grepl("NaNs",
            w$message)) | (W[1] == "sqrt") | (W[1] == "nlminb" &
            W[2] == "x.par") | (W[2] %in% c("m.SRRR.up", "m.SRRR.down",
            "m.SRSQD.up", "m.SRSQD.down") & grepl("model has NOT converged",
            w$message))) {
            invokeRestart("muffleWarning")
        }
    })
    chisq1 <- plyr::ldply(list(full = s.full, SRRR = s.SRRR,
        SRR = s.SRR, RR = s.RR, SQD = s.SQD), function(x) {
        chi <- -1
        if (!is.null(x)) {
            if (inspect(x, "converged") == TRUE)
                chi <- fitMeasures(x, "chisq")
        }
        return(chi)
    })

    chisq1 <- chisq1[chisq1[, 2] >= 0, ]
    if (nrow(chisq1) > 1) {
        chisq1$lag <- c(diff(chisq1[, 2], lag = 1), NA)
        if (any(chisq1$lag < 0, na.rm = TRUE)) {
            warning(paste0("There are convergence problems with model ",
                chisq1[which(chisq1$lag < 0), ".id"], ". Its chi-square value is higher than that of a nested model, which is theoretically not possible. Please inspect the results with care, using the compare()-function"))
        }
    }
    chisq2 <- plyr::ldply(list(full = s.full, SRRR = s.SRRR,
        SRSQD = s.SRSQD, SSQD = s.SSQD, SQD = s.SQD), function(x) {
        chi <- -1
        if (!is.null(x)) {
            if (inspect(x, "converged") == TRUE)
                chi <- fitMeasures(x, "chisq")
        }
        return(chi)
    })
    chisq2 <- chisq2[chisq2[, 2] >= 0, ]
    if (nrow(chisq1) > 1) {
        chisq2$lag <- c(diff(chisq2[, 2], lag = 1), NA)
        if (any(chisq2$lag < 0, na.rm = TRUE)) {
            warning(paste0("There are convergence problems with model ",
                chisq2[which(chisq2$lag < 0), ".id"], ". Its chi-square value is higher than that of a nested model, which is theoretically not possible. Please inspect the results with care, using the compare()-function"))
        }
    }
    modellist <- list(null = s.NULL, full = s.full, fullcluster = s.full.cluster, IA = s.IA,
        diff = s.diff, mean = s.mean, absdiff = s.absdiff, additive = s.additive,
        SQD = s.SQD, SRRR = s.SRRR, SRR = s.SRR, RR = s.RR, SSQD = s.SSQD,
        SRSQD = s.SRSQD, absunc = s.absunc, cubic = s.cubic,
        onlyx = s.onlyx, onlyy = s.onlyy, onlyx2 = s.onlyx2,
        onlyy2 = s.onlyy2, weak = s.weak, strong = s.strong)
    res <- list(models = modellist, SRSQD.rot = SRSQD.rot, SRRR.rot = SRRR.rot,
        LM = summary(lm.full), formula = formula, data = df,
        out.rm = out.rm, outliers = which(df$out == TRUE), DV = DV,
        IV1 = IV1, IV2 = IV2, IV12 = IV12, IV22 = IV22, IV_IA = IV_IA,
        W_IV1 = W_IV1, W_IV2 = W_IV2, IV13 = IV13, IV23 = IV23,
        IV_IA2 = IV_IA2, IV_IA3 = IV_IA3, r.squared = summary(lm.full)$r.squared)
    attr(res, "class") <- "RSA"
    return(res)
}
environment(RSA.akmod) <- asNamespace('RSA')

Spring 2018 Courses

Teaching, by andrew, October 26, 2017

Organizational Behavior
1st Year PMBA OB Core Course

This course presents a framework for thinking about how individual attributes and interpersonal skills serve as a foundation for effective leadership. As a context for developing these skills, the course focuses in particular on work in groups and teams. Through experiential exercises and classroom discussions, this course will enable you to gain deeper insights into your current strengths as a leader and to identify developmental opportunities for the future. There are two primary objectives:

  • Deepen your self-awareness by enhancing your insights into (1) your personal characteristics and attributes; (2) your interpersonal, social, and leadership skills; and, (3) your approach in working within groups and teams.
  • Improve your leadership effectiveness by enhancing your capacity to (1) identify your own leadership strengths and weaknesses and (2) understand how your assets and liabilities combine with others’ assets and liabilities in team-based work.

Dyadic data analysis

Research, by andrew, October 26, 2017

Knight, A. P., & Humphrey, S. E. (2019). Dyadic data analysis. In S. E. Humphrey and J. M. LeBreton (Eds.), The Handbook for Multilevel Theory, Measurement, and Analysis, pp. 423-447. Washington, DC: American Psychological Association.

Accompanying R functions for the social relations model: http://apknight.org/pdSRM.R

Abstract. Many foundational theories in the social sciences rely upon assumptions about dyadic interpersonal perceptions, behaviors, and relationships. This chapter provides a broad introduction to foundational concepts and techniques in analyzing dyadic data. The authors describe in detail one specific approach to dyadic data analysis—the social relations model—and provide software functions for conducting the analysis using multilevel modeling in R. The value of dyadic data analysis is illustrated through a discussion of prior publications that have used this approach. The authors also provide a step-by-step empirical example of how to use the social relations model with multilevel modeling in R, focused on dyadic trust in workgroups. The chapter concludes with a discussion of alternative approaches, beyond the social relations model, for analyzing dyadic data.

Innovations in unobtrusive methods

Research, by andrew, October 26, 2017

Knight, A. P. (2018). Innovations in unobtrusive methods. In A. Bryman and D. A. Buchanan (Eds.), Unconventional Methodology in Organization and Management Research, pp. 64-83. Oxford: Oxford University Press.

Abstract. Twenty years ago, engineer and computer scientist Rosalind Picard (1997, p.228) imagined a future in which ‘a financial analyst might combine his cell phone, pager, online stock reports, analysis software, and personal email agent into one computer that fits in a belt, watch and shirt pocket’.  Clearly the future is now.  An estimated 1.4 billion people owned a smartphone in 2013 – more than one fifth of the global population (Heggestuen, 2013).  By 2020, that proportion is expected to rise to approximately 70 percent (Ericsson, 2015).  And smartphones are just the tip of the iceberg, as a proliferation of internet-connected devices expands the linkages among humans, computers, and networks.  Consider just a few of the devices released recently.  Glasses developed by companies like Google and Snap enable users to capture and share multimedia content in real-time; wristbands like those developed by Fitbit, Apple, and Samsung facilitate fitness tracking, payments, and more.

The ubiquity of connected devices (Swan, 2012) – and the metrics that they unobtrusively capture – has led data to become increasingly central to the global economy.  Companies have integrated novel unobtrusive data streams into their business models and operations (e.g. Walker, 2012; Wilson, 2013).  These data streams can elucidate consumer preferences and responses to advertising, enhance human resource practices, and improve collaboration networks – to name just a few publicized applications.

Much like new data streams have enriched contemporary businesses, innovative unobtrusive methods hold great promise for researchers who study organizational functioning (Tonidandel et al., 2016).  The idea that researchers can benefit from using unobtrusive methods is certainly not new.  More than half a century ago, Webb and colleagues (1966) implored researchers in their classic book Unobtrusive Measures to use a more diverse set of data streams in their work, noting that, ‘Today the dominant mass of social science research is based upon interviews and questionnaires.  We lament this overdependence upon a single fallible method’ (pp.1-2).  Notwithstanding a steady drumbeat of pleas over the years for researchers to use unobtrusive methods (e.g. Hill et al., 2014; Webb and Weick, 1979), survey methods continue to dominate the literature, especially in organizational behaviour, and researchers still often rely on a single data source (Podsakoff et al., 2012; Scandura and Williams, 2000).

The purpose of this chapter is to describe a new suite of unobtrusive methods, such as the traces that people leave throughout the digital world as they search the Internet, post content on social media, and navigate an increasingly digitally-connected physical world.  These methods, which did not exist when Webb and colleagues published their book, make it easier and cheaper for researchers to use unobtrusive methods than ever before.  As a result, we social science researchers have fewer and fewer excuses for relying on a single source of data, obtrusively acquired, in empirical studies.

Fall 2017 Courses

Teaching, by andrew, August 18, 2017

Leadership Development
2nd Year MBA OB Core Course

This course builds upon the material from the 1st Year OB Core (OB 5620, Foundations for Leadership Effectiveness) and, importantly, from your time so far at Olin and during your summer work experiences. The focus of the course is on the attributes, behaviors, and tendencies of effective leadership. There are two primary objectives:

  • Gain new insights into your own beliefs and expectations regarding what constitutes effective leadership in groups, teams, and organizations. You will accomplish this through a mixture of classroom discussion, case analysis, and self assessment.
  • Learn about your own strengths and weaknesses in leading others. You will accomplish this in the classroom through controlled experiential exercises, which will be the basis for feedback from your peers. You will also reflect on your strengths and weaknesses as revealed in critical incidents from your summer work experiences.

Foundations for Effective Leadership / Organizational Behavior
1st Year MBA and PMBA OB Core Course

This course presents a framework for thinking about how individual attributes and interpersonal skills serve as a foundation for effective leadership in small groups and teams. Through experiential exercises and classroom discussions, this course will enable you to gain deeper insights into your current strengths as a leader and to identify developmental opportunities for the future. There are two primary objectives:

  • Deepen your self-awareness by enhancing your insights into (1) your personal characteristics and attributes; (2) your interpersonal, social, and leadership skills; and, (3) your approach in working within groups and teams.
  • Improve your leadership effectiveness by enhancing your capacity to (1) identify your own leadership strengths and weaknesses and (2) understand how your assets and liabilities combine with others’ assets and liabilities in team-based work.

Identity Literacy: An Introduction to Cultural Competence in a Diverse World
University 1st Year Undergraduate Course

This course provides students the opportunity to explore how identities form, intersect, and are negotiated within some of our nation’s most visible and controversial contemporary moments. In grappling with how identities are constructed and contested, you will be exposed to the processes of reflection, vocabulary, and concepts that will prepare you to engage with a variety of identity groups. You will also be introduced to the ways in which political, economic, and institutional structures interact with identity.

This course assumes that no matter what particular ideologies and attitudes you embody, you will encounter people whose identity differs from your own, and will need the knowledge and skills that are essential to having responsible, informed, and successful interactions. In fact, each course participant is likely to have multiple dominant and non-dominant identities that intersect: a student may be Native American but also part of the Christian majority, while another student may be from abroad but nonetheless have male privilege in the U.S. The course consequently is not intended to promote any particular identity as more deserving of understanding, but instead presents several of them as opportunities for you to develop a general framework for understanding difference, while reflecting on your own identity and acquiring the skills necessary for successful interaction in diverse social and professional settings.

In every profession, cognizance of the difference identity can make in how people interact with the world (and how the world interacts with people) is essential to being a good citizen, but also an effective member of a team, leader, or knowledgeable practitioner in one’s chosen field. What difference does understanding gender identity and expression make in the questions you ask as a doctor, or presumptions people make in everyday conversation? How important is it to know the difference class and race has made in the development of the cities we inhabit, and the choices we make about how we interact with the city? How can learning how to have dialogues (as opposed to debates) about complex social issues help us become better listeners, citizens, and professionals in whatever spaces we inhabit?

Finally, enrollment in this course is based on residential floors because “living the curriculum” enriches learning, and allows you to apply what you’ve learned to the relationships you build with your peers. Gaining knowledge about varied identities together we hope will enhance the mutual respect you have for each other, build an inclusive environment, and enhance your student experience at Washington University.

Organizational affective tone

Research, by andrew, February 16, 2017

Knight, A. P., Menges, J. I., & Bruch. H. (2018). Organizational affective tone: A meso perspective on the origins and effects of consistent affect in organizations. Academy of Management Journal, 61, 191-219.

Abstract. Grounded in an open systems perspective, we build and test new theory about how the kinds of industries in which an organization participates influences organizational affective tone and connects to workforce strain. We propose that the more an organization’s activities lie in consumer-centric industries (e.g., service, retail), the more positive and less negative the organization’s affective tone. We connect consumer-centric industry participation and affective tone by explaining how personnel policies and organizational structure generate and sustain consistent positive and negative affect throughout an organization. Additionally, we examine the effects of organizational affective tone on workforce strain. The results of a survey-based study of 24,015 human resource managers, top management team members, and employees of 161 firms largely support our predictions. We discuss the implications of considering macro contextual factors for understanding affect in organizations.

Resources and relationships in entrepreneurship

Research, by andrew, January 3, 2017

Huang, L., & Knight, A. P. (2017). Resources and relationships in entrepreneurship: An exchange theory of the development and effects of the entrepreneur-investor relationship. Academy of Management Review, 42, 80-102.

Abstract. We develop a theoretical model, grounded in exchange theory, about the process through which relationships between entrepreneurs and investors develop and influence the growth of new ventures. Our theory highlights the multifaceted relationships that entrepreneurs and investors share—comprising both affective and instrumental dimensions—and the bidirectional exchanges of social and financial resources that build these relationships over time. An exchange theory perspective sheds light on the emergence of different patterns of relationship development over time and how different kinds of resource exchange contribute to new venture growth, contingent on the core problems that a venture faces at a given stage of development. We discuss implications of an exchange perspective on resources and relationships in entrepreneurship for theory, research, and practice.