This paper proposes a new approach that can discriminate risky software development projects from smoothly or satisfactorily going projects and give explanation for the risk. We have already developed a logistic regression model which predicts whether a project becomes risky or not. However, the model returned the decision with the calculated probability only. Additionally, a formula was constructed based on the risk questionnaire which includes 23 questions. We therefore try to improve the previous method with respect to accountability and feasibility. In new approach, we firstly construct a new risk questionnaire including only 9 questions (or risk factors) Q1, Q2, \cdot\cdot\cdot, Q9, each of which concerns with the project management. We then apply multiple regression analysis to actual project data PJ1, PJ2, \cdot\cdot\cdot, PJ32, and clarify a set of factors which contributes essentially to estimate the relative cost error and the relative duration error, respectively. We then apply the constructed formulas to another project data PJ33, PJ34,\cdot\cdot\cdot, PJ40. The analysis results show that both the cost and duration of risky projects are estimated fairly well by the formulas. We can thus confirm that our new approach is applicable to software development projects in order to discriminate risky projects from appropriate projects and give reasonable explanations for the risk.